A Review of the Comparative Morphology of Extant Testudinoid Turtles (Reptilia: Testudines)

Transcription

1 2004 Asiatic Herpetological Research Vol. 10, pp A Review of the Comparative Morphology of Extant Testudinoid Turtles (Reptilia: Testudines) WALTER G. JOYCE 1 AND CHRISTOPHER J. BELL 2 1 Department of Geology and Geophysics, Yale University, New Haven, CT walter.joyce@yale.edu 2 Department of Geological Sciences, The University of Texas at Austin, Austin, TX cjbell@mail.utexas.edu Abstract. - With an expansive geographic distribution, an excellent fossil record, and over 140 recognized extant species, testudinoid turtles constitute one of the most diverse and widespread clades of turtles. The current understanding of the distribution of morphological characters among testudinoid turtles is poor. Improved knowledge will help to facilitate accurate identification of fossil remains, and to provide a reliable morphological data set for phylogenetic analyses. We provide a critical review of skeletal and scute characters commonly utilized in previous systematic analyses of Testudinoidea. Description and illustration of character states, discussion of their distribution within Testudinoidea, and polarity determinations for 93 characters are provided. Our preliminary results indicate that ontogenetic changes in skeletal structure are an important source of variation within Testudinoidea. Sexual variation, ontogenetic variation, and intra- and inter-population variation are inadequately documented for most testudinoid taxa. Furthermore, data matrices of morphologic characters in the existing literature must be carefully reconsidered. Previously published morphologic data provide reasonably strong support for the monophyly of 'Testudinidae.' Strong morphologic support for a monophyletic 'Emydidae' is lacking, and 'batagurid' monophyly has not been rigorously tested in the literature. Because a new research cycle centered on testudinoid phylogeny is now under way, it is essential to critically re-examine the underlying assumptions and working hypotheses that have governed this field of study over the last 20 years. Key words. - Testudines, Testudinoidea, Testudinidae, Emydidae, Bataguridae, Geoemydidae, morphology, systematics Introduction Pond turtles and land tortoises (collectively, Testudinoidea) form one of the largest and most widespread clades of living turtles, with more than 140 extant species and an almost worldwide distribution. The discovery and description of many new fossil testudinoids in the last half century, combined with the emergence and ascendancy of molecular techniques in systematics, provide new opportunities to explore the evolutionary history of the group in unprecedented detail. Concomitant with the appearance of these new data sets and analytical techniques comes an increasing appreciation for conservation efforts to preserve these turtle lineages and help to secure their future in the face of increasing human predation and habitat encroachment. This is true especially for the Asian representatives of this clade (e.g., van Dijk et al., 2000) but also is relevant at a more generalized and inclusive level (e.g., Rhodin, 2000). Our recent attempts to diagnose fossil testudinoids reliably and to place them within a phylogenetic context led to the recognition that a critical re-evaluation of morphological data and purported synapomorphies for the subclades of testudinoid turtles is desirable. A more thorough understanding of morphological data sets will provide not only a means by which molecular trees may be independently assessed, but also will form an essential foundation for diagnosing and interpreting fossil specimens. This in turn will facilitate the integration of fossil taxa into future systematic analyses, and will enhance our understanding of the paleobiogeography and divergence times of extant lineages. The recent flurry of published works appears to represent the beginning of a new research cycle (sensu Kluge, 1991) in testudinoid systematics. We suggest that an important part of this cycle will be a critical re-examination of the working hypotheses that have governed testudinoid systematics since the publication of McDowell s (1964) seminal work on the group. A key component of this will be the assessment of fundamental, often unstated, assumptions that underlie current hypotheses of relationship. Our contribution to this research cycle is the first critical reappraisal of morphological characters applied to testudinoid systematics since the work of Hirayama (1985). The emerging improvement in our understanding of testudinoid relationships based on molecular sequence data will certain by Asiatic Herpetological Research

2 Vol. 10, p. 54 Asiatic Herpetological Research 2004 ly result in numerous new questions (e.g., regarding paleobiogeography, the timing of sequence and evolutionary divergences, and the evolution of morphological adaptations) that will demand a clearer understanding of testudinoid morphology. The purpose of this paper is to present a preliminary revision and discussion of the morphological characters previously utilized in investigations of testudinoid systematics. Our goal here is not to produce a phylogenetic hypothesis (indeed, we deliberately eschew such a production), but rather to evaluate the morphological data that have been, and will be, used to generate such hypotheses. To enhance our discussion and facilitate improved communication about testudinoid morphology, we provide illustrations of all characters states we discuss. Abbreviations. - Institution and collection abbreviations: CAS, California Academy of Sciences, San Francisco, California; CM, Carnegie Museum of Natural History, Pittsburgh, Pennsylvania; FMNH, Field Museum of Natural History, Chicago, Illinois; IVPP, Institute of Vertebrate Paleontology and Paleoanthropology, Beijing, China; KU, The University of Kansas Natural History Museum, Lawrence, Kansas; LMNH, Louisiana Museum of Natural History, Baton Rouge, Louisiana; MCZ, Museum of Comparative Zoology, Harvard University, Cambridge, Massachusetts; TNHC, Texas Natural History Collections, Texas Memorial Museum, Austin, Texas; TUMNH, Tulane University Museum of Natural History, New Orleans, Louisiana; YPM, Yale Peabody Museum, New Haven, Connecticut. Abbreviations used in figures: AB, abdominal scute; af, articular facet; an, angular; bo, basioccipital; CE, cervical scute; co, costal bone; ent, entoplastron; epi, epipterygoid; fdm, foramen dentofaciale majus; fr, frontal; fpp, foramen palatinum posterius; HU, humeral scute; hyo, hyoplastron; hypo, hypoplastron; ju, jugal; MA, marginal scute; mx, maxilla; ne, neural bone; pa, parietal; pal, palatine; PEC, pectoral scute; pf, prefrontal; PL, pleural scute; pm, premaxilla; po, postorbital; pt, pterygoid; qj, quadratojugal; qu, quadrate; VE, vertebral scute; vf, vomerine foramen; vo, vomer. Material and Methods We examined 309 testudinoid specimens representing 93 species, but focused our efforts on 46 representative species. The list of specimens examined is provided in Appendix 1. Turtle shell nomenclature follows Zangerl (1969) and cranial nomenclature follows Gaffney (1972). Of the 46 focal species, most were recognized as valid species by Ernst and Barbour (1989), with the exception of texana, which they placed within concinna. Generic allocations for testudinoid species varied widely over the last 50 years and are subject to differing opinions today, particularly because the monophyly of many testudinoid genera remains untested. We consequently suppress the use of generic names wherever possible and use species epithets only. This procedure also has the advantage of precisely associating observations with species only instead of higher taxonomic categories. Most extant turtles have distinct species names, but among those turtles discussed in this review the species epithets insculpta, nelsoni, oculifera, ornata, and platynota each appear twice (insculpta under Glyptemys and Carettochelys; nelsoni under Pseudemys and Terrapene; oculifera under Graptemys and Psammobates; ornata under Pseudemys and Terrapene; platynota under Geochelone and Notochelys). For clarity in these instances, we indicate our usage with a single-letter generic abbreviation. A complete list of all currently recognized testudinoid species and all outgroup species used herein is provided in Appendix 2 together with a list of their various generic assignments used in the last 50 years. Our use of the classic higher categories is always restricted to their phylogentic crown. Emydinae (sensu McDowell, 1964) are also referred to as Emydidae, emydids, or North American pond turtles; Batagurinae (sensu McDowell, 1964) as Bataguridae, batagurids or Asian pond turtles; and Testudinidae (sensu McDowell, 1964) as testudinids or (land) tortoises. We make no a priori assumptions of monophyly for any of these categories, and retain single quotations around these names throughout the text to emphasize our uncertainty. We attempted to examine most significant morphological characters commonly utilized in systematic studies of testudinoids, but the majority of our observations concern the skeletal system and scute characters. Almost all characters were derived from the literature. Major sources for each category were: Batagurinae (Hirayama, 1985; McCord et al., 1995; Yasukawa et al., 2001); Testudinidae (Crumly, 1982, 1985, 1994); and Emydinae (Gaffney and Meylan, 1988; Burke et al., 1996). Additional characters were also found in Mlynarski (1976), Shaffer et al. (1997), and other sources cited in the character discussions. With few exceptions, morphological features were examined on specimens themselves; evaluations based on previously published literature are indicated where applicable. Sexual dimorphism, ontogenetically influenced polymorphisms, and geographic variation in morphology are not well explored in testudinoid turtles. These areas are in need of much more research. A full exploration of

3 2004 Asiatic Herpetological Research Vol. 10, p. 55 such variation is beyond the scope of this work, but we are able to make some preliminary observations regarding morphological change through ontogeny in some anatomical systems. Polarizing characters with the help of outgroups proved to be a difficult task, mostly because all relevant extant sister taxa are highly specialized after more than 65 million years of independent evolution. Furthermore, hypotheses of the systematic relationships of the major groups of cryptodires reveal a highly unstable picture (e.g., Bickham, 1981; Gaffney, 1975, 1985; Gaffney et al., 1991; Shaffer et al., 1997) making it impossible to make any a priori decisions regarding the succession of outgroups. We consequently assessed polarity for most characters by examining select outgroup taxa and the ingroup taxa. Where polarity is not clear from outgroup comparison, we sometimes relied upon ingroup commonality. To allow full transparency, we discuss every polarity decision at the end of each character description. Outgroup taxa include the cryptodires caretta, odoratus, serpentina, and spinifera, and the pleurodires gibba, siebenrocki, subglobosa, and subrufa. For a number of characters, especially of the shell, neither ingroup nor outgroup analysis of extant taxa proved useful. In these instances, polarity was based on literature descriptions of the lindholmemydid taxa Gravemys, Lindholmemys, and Mongolemys. These Cretaceous, Asian, fossil taxa are not well described in the literature, but sufficient material and description exists to use these taxa to help polarize character states (e.g., Khosatzky and Mlynarski, 1971; Sukhanov, 2000; Danilov and Sukhanov, 2001). The group may not be monophyletic, but putative members currently are hypothesized to sit along the phylogentic stem of Testudinoidea (Danilov and Sukhanov, 2001). We purposefully did not use the fossil taxon Echmatemys as an outgroup taxon (Hirayama, 1985), because its phylogenetic position outside of Testudinoidea or even Batagurinae is not sufficiently demonstrated. All figures were produced using digital photography and processed using Adobe Photoshop. Images were digitally enhanced using the burn and burnish tools and the unsharp mask filter option. Taxonomic and Systematic Background. - Despite the increased attention directed towards testudinoids by scientists, hobbyists, and nonprofessional enthusiasts in the last thirty-five years, our collective conceptualization of the higher-level (beyond the specific and generic) systematics within this clade remained virtually unchanged since the work of McDowell (1964). The various taxonomies in current use owe their existence in large part to historical contexts that are not well appreciated by many authors. A brief summary is given here. During the second half of the 19th century, a number of attempts were made to work out higher-level testudinoid relationships and to apply taxonomic conventions that were designed (to a greater or lesser extent) to communicate conceptualizations of these relationships. In his synopsis on the turtles of North America, Agassiz (1857) united all pond turtles into the Emydoidae and subdivided this group into a three monotypic subfamilies (Deirochelyoidae for reticularia, Evemydoidae for blandingii, Cistudinina for T. ornata and carolina) followed by the subfamilies Clemmydoidae (for G. insculpta, guttata, marmorata, and muhlenbergii) and Nectemydoidae (for those species currently placed in the genera Pseudemys, Trachemys, Graptemys, Malaclemys and Chrysemys). Most land tortoises were placed in Testudinidae by Theobald (1868); he also included all leaf turtles and tortoises (e.g., amboinensis, emys, dentata, grandis, tricarinata) in Geoemydidae, and an eclectic group of aquatic turtles, including megacephalum, serpentina, kinosternids, and all remaining testudinoids, in Emydidae. Subsequently, all land tortoises (including emys and impressa) were united in the Testudinidae by Gray (1870). Those species currently placed in Pseudemys and Trachemys were assigned by Gray (1870) to the Pseudemydae; the Asian taxa baska, borneoensis, thurjii, kachuga, and ocellata were assigned to the Bataguridae, and all hinged pond turtles, true terrapins, and snail-eating pond turtles to the Holarctic families Cistudinidae, Emydidae, and Malaclemmydae, respectively. Despite these early attempts, most subsequent authors (e.g., Boulenger, 1889; Siebenrock, 1909; Lindholm, 1929; Smith, 1931; Bourret, 1941) ignored (or were unaware of) these works and simply divided all testudinoid turtles into two speciose subgroups: tortoises (Testudinidae or Testudininae) and pond turtles (Emydidae or Emydinae). This situation remained static for nearly 100 years until the comprehensive and influential work of McDowell (1964). He not only divided all known pond turtles into several species complexes ( Hardella, Batagur, Orlitia, Geoemyda, Chrysemys, Deirochelys, and Emys complexes), but also concluded that pond turtles can be divided clearly into two subgroups, the predominantly North American Emydinae and the Asian and central American Batagurinae. Furthermore, McDowell (1964) reasoned that tortoises are not the sister group of pond turtles, but rather were likely derived from a batagurine ancestor. These conclusions were later corroborated by the first, and to date only, comprehensive morphological cladistic analysis of batagurine systematics (Hirayama, 1985). The influence of McDowell s (1964) work is best understood when considering its continuous impact on

4 Vol. 10, p. 56 Asiatic Herpetological Research 2004 subsequently proposed phylogenies. Despite differences of opinion regarding generic- and species-level systematic arrangements, virtually all major synthetic works in the last thirty years followed McDowell s (1964) subdivision of pond turtles into the Batagurinae and Emydinae (e.g., Mlynarski, 1976; Pritchard, 1979; Ernst and Barbour, 1989), even though Batagurinae may best be regarded as a paraphyletic taxon (McDowell, 1964; Hirayama, 1985). The fundamental division proposed by McDowell is also reflected in more recent studies centered on using various molecular techniques to elucidate phylogeny, the majority of which dealt with treatments of in-group relationships within one or the other of McDowell s groups (Sites et al., 1984; Bickham et al., 1996; Carr and Bickham, 1986; Wu et al., 1999; McCord et al., 2000; Feldman and Parham, 2002; Honda et al., 2002; Iverson et al., 2002; Stephens and Wiens, 2003). Admittedly, the list of autapomorphic characters compiled by Crumly (1985) for the Testudininae compellingly corroborates the hypothesis of tortoise monophyly. However, most characters that currently unite Emydinae or Batagurinae + Testudininae seem to support these groupings weakly, because the derived states typically are not found within all species of the ingroup and commonly also are observed in species of the alleged sister group (e.g., Hirayama, 1985; Gaffney and Meylan, 1988). In addition, several of the characters that purportedly distinguish Batagurinae + Testudininae from the Emydinae probably should be considered primitive for the entire group (Gaffney and Meylan, 1988). Even if some characters do successfully unite a group, monophyly is not established until the involved characters are demonstrated to be derived within Testudinoidea. Furthermore, the simple demonstration of monophyly for a given group does not automatically imply that it must be the sister to the remaining taxa. For instance, Emydinae may be monophyletic, but monophyly does not necessarily demand that Emydinae be regarded as the sister to Batagurinae + Testudininae. It is at least plausible that Emydinae is situated within Batagurinae, a possibility that is not adequately explored and tested in the literature. Similarly, most of the groupings considered by Gray (1855, 1870) and Agassiz (1857) were not discussed in recent literature, even though they might be valid. For instance, given the considerable list of morphological similarities that are shared by hinged turtles of the New and Old World (e.g., development of a plastral hinge, reduction of posterior neural elements, fusion of the femoral trochanter, great reduction of the temporal arch) perhaps Gray (1870) was truly visionary in uniting these turtles as the Cistudinidae. Only a global cladistic analysis with no a priori assumption regarding internal relationships can evaluate these alternatives and produce testable results. It is toward this end that we offer our critical reappraisal of morphological characters in testudinoids. Results and Discussion Cranium (1) Shape of the fissura ethmoidalis; 0 = narrow or closed, keyhole-shaped, Fig. 1; 1 = very wide, Fig. 2 (modified from Crumly, 1982, 13; Hirayama, 1985, 1; McCord et al., 1995, 5). The general configuration of the fissura in emydids and batagurids is keyhole-shaped (McDowell, 1964). Different proportions and widths are apparent (especially in the ventral part of the fissura), and were scored by Hirayama (1985) and Crumly (1982) as discrete character states. Our survey of many taxa reveals morphological intermediates, and the expression of various states appears to have an ontogenetic component in which younger individuals exhibit a relatively larger fissura, which corresponds to a less-ossified nasal cavity. However, a rather significant morphological gap can be observed between tortoises and pond turtles. For the purpose of this review, we lumped Hirayama s (1985) states into our state 0, and Crumly s (1982) into our state 1. The scoring presented by Hirayama (1985) and McCord et al. (1995) permitted phylogenetic resolution within batagurids, and that of Crumly (1982) within tortoises. Our revised scoring permits support only for the hypothesis of a monophyletic Testudinidae. Polarity: Pleurodires lack a defined fissura ethmoidalis. A keyhole-shaped fissura ethmoidalis is present in spinifera, odoratus, caretta, serpentina, and Mongolemys, and this condition is considered primitive for testudinoids. (2) Medial inflection of the inferior descending processes of the frontal; 0 = absent, or very small, Fig. 3; 1 = present, well-developed, medial contact present or almost present, Fig. 4 (modified from Hirayama, 1985, 2). In most turtles, a gutter (the sulcus olfactorius) is formed along the ventral surface of the frontals. This gutter transmits the olfactory nerve. The lateral rims of the sulcus sometimes form processes that descend ventromedially to surround the nerve from below (McDowell, 1964). According to Hirayama (1985) these processes are well-developed, or are in contact medially, in ocellata and hamiltonii. We confirm the presence of well-developed processes in these taxa and add petersi, N. platynota, and all sampled tortoises to the list. We recommend that this character not be subdivided into additional character states, because the descending processes of the frontals grow larger through ontogeny.

5 2004 Asiatic Herpetological Research Vol. 10, p. 57 Polarity: A medial inflection is absent in all outgroups and the vast majority of the ingroup. We consider its presence to be derived. (3) Frontal contribution to the orbital rim; 0 = present, no prefrontal/postorbital contact on dorsal surface, Fig. 5; 1 = absent, frontal excluded from orbital rim by prefrontal/postorbital contact, Fig. 6 (modified from Crumly, 1982, 17; Hirayama, 1985, 3; Shaffer et al., 1997, 97; Yasukawa et al., 2001, 1). Three states for this character were scored by Hirayama (1985) and Yasukawa et al. (2001): frontal contribution always or usually present, frontal sometimes excluded from orbital rim, and frontal always excluded from orbital rim. Our sample size for many taxa does not permit a reliable assessment of intraspecific variation in this character, and thus our initial scores differed for some taxa from those of Hirayama (1985). We add petersi and N. platynota to the list of taxa in which the frontal appears always to be excluded. For those taxa that sometimes exclude the frontals, we confirm this polymorphic condition in crassicollis, and add agassizii and annulata. Our sample was too small to confirm the reported polymorphic condition in amboinensis, and pulcherrima by Hirayama (1985) and Yasukawa et al. (2001), but we scored these taxa as polymorphic based on their observations. We also followed Crumly (1982) by coding pardalis as polymorphic, even though we were not able to observe this in our sample. Our coding differs from that of Shaffer et al. (1997) for Heosemys and reevesii. In their analysis, they used spinosa (in which the frontal contributes to the orbital rim), but we used grandis (in which it does not). In both specimens of reevesii available to us, the frontal clearly does not participate in the orbital margin. Given the contrary statement by Shaffer et al. (1997), reevesii may be polymorphic for this character. Polarity: The frontal participates in the orbital rim in pleurodires and spinifera, it is excluded in odoratus and caretta, and it is polymorphic in serpentina. No pattern is apparent within the ingroup. Given that the frontal clearly contributes to the orbital rim in Mongolemys, we consider its absence to be derived. (4) Contact between jugal and pterygoid; 0 = present, medial process of jugal well-developed and touching the pterygoid, Fig. 7; 1 = absent, medial process reduced, Fig. 8 (modified from Hirayama, 1985, 11, 12; McCord et al., 1995, 3; Burke et al., 1996, 23; Yasukawa et al., 2001, 4, 5). The jugal of most testudinoid turtles is expanded at its ventral end to form a medial process that contacts the pterygoid medially (McDowell, 1964). Presence or absence of the medial process, and presence or absence of a medial contact with the pterygoid were treated as two characters by Hirayama (1985) and Yasukawa et al. (2001). The scoring for the two characters appears to be redundant and we followed the recommendation of Gaffney and Meylan (1988) by combining them. We confirm the loss of a medial contact between the jugal and the pterygoid in galbinifrons, flavomarginata, and mouhotii (Hirayama, 1985; Yasukawa et al., 2001), but we found this condition to be polymorphic in spengleri (also reported by McCord et al., 1995), and in trijuga. Our observations are concordant with those of Burke et al. (1996). Polarity: A contact between the jugal and pterygoid in present in spinifera, odoratus, and serpentina, but is absent (although the two bones closely approach one another) in caretta. We conclude that the contact between the two bones is the primitive condition for testudinoids and that their separation is derived, a conclusion also reached by Hirayama (1985). Our polarity determination is opposite that used by McCord et al. (1995), who mistakenly claimed to have derived their polarity assessment from Hirayama (1985). (5) Contact between jugal and palatine; 0 = absent, Fig. 9; 1 = present, Fig. 10 (Gaffney and Meylan, 1988, F5.4). The presence of a contact between the medial process of the jugal and the palatine was used previously in support of a monophyletic Deirochelyinae (Gaffney and Meylan, 1988). We confirm the formerly observed distribution of this character within emydids with the exception of reticularia, which does not exhibit a contact. A contact is present in numerous batagurids, such as borneoensis, reevesii, and hamiltonii, but was absent in all examined members of Testudinidae. Polarity: A contact between the jugal and the palatine is present in caretta, odoratus, and pleurodires, but is absent in spinifera and serpentina. A contact is absent in Mongolemys. We consequently consider its presence to be derived. (6) Contact of the epipterygoid with the jugal; 0 = clearly absent, Fig. 11; 1 = present, or almost present, epipterygoid forms a long lateral process that approaches the jugal, Fig. 12 (Gaffney and Meylan, 1988, F8.1; Shaffer et al., 1997, 106). According to Gaffney and Meylan (1988) the epipterygoid and the medial process of the jugal approach one another or are in contact in reticularia and the various species they included in Pseudemys and Trachemys. They also noted a contact between these two elements in species they classified in Graptemys, but the condition in those taxa was interpreted to be a result of

6 Vol. 10, p. 58 Asiatic Herpetological Research 2004 the medial expansion of the jugal and not a lateral expansion of the epipterygoid, and consequently was regarded as non-homologous (Gaffney and Meylan, 1988). We confirm the contact or near contact of these two elements in reticularia, decorata, scripta, alabamensis, P. nelsoni, rubriventris, texana, flavimaculata, geographica, kohnii, nigrinoda, G. oculifera, ouachitensis, and versa, and also report it in picta and terrapin. Contact was clearly absent in the specimens of barbouri, ernsti, and gibbonsi we examined. We made no assessments of homology, but accept any contact between these two elements as the derived state (as was done by Shaffer et al., 1997). Among batagurids, we also found a close approach in reevesii. Polarity: A contact, or near contact, between the epipterygoid and the jugal is absent in all outgroups and the vast majority of the ingroup. We consider its presence to be derived. (7) Contact of the inferior process of the parietal with the medial process of the jugal; 0 = absent, Fig. 13; 1 = present, Fig. 14 (Hirayama, 1985, 13). Our coding differs significantly from that of Hirayama (1985). In reevesii, N. platynota, and bealei we found no contact between the parietal and jugal, although these were the only three taxa in which Hirayama (1985: table 2) scored it to be present. However, we found a pronounced contact between these two elements in subtrijuga, a species scored by Hirayama (1985: table 2) as lacking such a contact, but shown on his tree (Hirayama, 1985: fig. 2) as a unique batagurid feature convergent with some emydids. Among emydids, a well-developed contact occurs in barbouri and other broad-headed species currently classified in Graptemys. Polarity: There is no contact between the inferior process of the parietal and the medial process of the jugal in all outgroups and the vast majority of the ingroup. A contact is considered to be the derived condition. (8) Contact of the inferior process of the parietal with the maxilla; 0 = absent, Fig. 13; 1 = present, Fig. 14 (Hirayama, 1985, 14). Our coding differs from that of Hirayama (1985). According to his character matrix (table 2) a contact should be present between the inferior process of the parietal and the maxilla in reevesii and mouhotii, but his cladogram (fig. 2) indicated that the presence of a contact should be regarded as a uniquely derived autapomorphy of subtrijuga. We found no contact in reevesii or mouhotii. Of the testudinoid species we examined, subtrijuga is the only one that shows this feature. Polarity: There is no contact between the inferior process of the parietal and the maxilla in all outgroups. Its presence is considered to be derived. (9) Extent of quadratojugal; 0 = quadratojugal well developed, firmly attached to jugal, Fig. 15; 1 = quadratojugal present, contact lost with jugal, Fig. 16; 2 = quadratojugal so heavily reduced that it appears to be absent in many skeletal specimens, Figs. 17, 18 (modified from Hirayama, 1985, 16; Shaffer et al., 1997, 47; Burke et al., 1996, 21; McCord et al., 1995, 6; Yasukawa et al., 2001, 7, 8). Variation in the structure of the temporal region of turtles was discussed in detail by Zdansky (1924) and comments specific to testudinoids were provided by Zangerl (1948) and McDowell (1964). We originally scored the reduction of the quadratojugal as three different characters: loss of contact with the jugal, loss of contact with the squamosal, and the apparent loss of the quadratojugal. All five logically possible combinations were observed, but in most testudinoid turtles the temporal arch is so slender that the contact between the quadratojugal and squamosal is commonly reduced to a sliver that would have to be scored as just barely present or just barely absent. We therefore abandoned our efforts to evaluate the contact between the quadratojugal and squamosal. Our observations generally agree with those of McDowell (1964), Hirayama (1985), Burke et al., (1996), McCord et al. (1995), and Yasukawa et al. (2001). We purposefully avoid addressing the apparent lack of a quadratojugal in many species as an absence, because previous work by Zdansky (1924) showed that the quadratojugal of some batagurids is so poorly ossified and connected to the surrounding elements that it tends to be lost in skeletal specimens (Figs. 17, 18). An example of this problem can be found among the many conflicting statements made regarding the presence of this element in N. platynota (e.g., Smith, 1931; Bourret, 1941; McDowell, 1964; Ernst and Barbour, 1989). Polarity: The quadratojugal is present and firmly attached to the jugal in all outgroups, with the exception of chelids. Its reduction is considered to be derived. (10) Contribution of jugal to the rim of upper temporal emargination (Hirayama, 1985, 15); 0 = absent, Figs. 19, 20; 1 = present, Fig. 21. Participation of the jugal in the rim of the upper temporal emargination was reported previously in hamiltonii and ocellata (Hirayama, 1985). We confirm its presence in both species, but in one of the hamiltonii specimens we examined (MCZ ) the jugal forms a significant part of the rim only on one side of the skull; on the other side, which appears abnormal and likely represents a teratology, it does not. In all other species

7 2004 Asiatic Herpetological Research Vol. 10, p. 59 available to us, the jugal does not participate in the rim. In subtrijuga, the jugal is excluded from the upper temporal emargination by narrow extensions of the postorbital and quadratojugal (Fig. 20). Polarity: The jugal participates in the upper temporal rim of spinifera, but it is excluded in odoratus, caretta, serpentina, Mongolemys, and most pleurodires. We consider the participation of the jugal in the rim of the upper temporal emargination to be the derived condition within testudinoids. (11) Contact between the quadratojugal and the articular facet of the quadrate; 0 = absent, Fig. 22; 1 = present, quadratojugal sends a process ventrally along the rim of the cavum tympani and touches the lateral edge of the articular facet, Fig. 23 (modified from Hirayama, 1985, 17). The original character definition (Hirayama, 1985, character 17) is inappropriate, because the jugal does not contact the articular surface of the quadrate in any turtle except for madagascariensis and dumerilianus (Gaffney and Meylan, 1988). However, because Hirayama indicated in his tree that the only batagurid taxon to exhibit this character is subtrijuga, we assume that he was referring to a contact between the quadratojugal and the articular facet of the quadrate, a characteristic of subtrijuga only among testudinoids. A contact between the two elements was reported previously for reevesii in the character matrix published by Hirayama (1985), but we conclude that this must be a publishing error, because it stands in conflict with his tree. In the specimens of reevesii available to us, there is no contact. It is possible that the scoring for subtrijuga and reevesii were flipped, at least in part, in the Hirayama (1985) matrix (in which the taxa were listed next to one another). Polarity: A contact between the quadratojugal and the articular surface of the quadrate is absent in spinifera, but present in odoratus, caretta, and serpentina, and consequently could be considered primitive. However, based on ingroup commonality and the absence of a contact in Mongolemys, we consider a contact to be derived for Testudinoidea. (12) Contact between quadratojugal and maxilla; 0 = absent, Fig. 22; 1 = present, Figs. 23, 24 (Hirayama, 1985, 18). According to Hirayama (1985), among batagurids a contact between the quadratojugal and maxilla is only present in subtrijuga and reevesii. We did not find a contact in our specimens of reevesii, but confirm its presence in subtrijuga. Gaffney and Meylan (1988) listed a contact between the quadratojugal and maxilla as a synapomorphy for Platysternina {megacephalum + Chelydropsis } and as an independently evolved synapomorphy for Kinosternidae, whereas Shaffer et al. (1997) noted a contact to be present in Kinosternidae, C. insculpta, and megacephalum. Our observations confirm the presence of a contact in all of these extant groups. Among testudinoids, subtrijuga is unique in having an extensive contact in lateral view (Fig. 23). In several emydids, a contact is present on the inside of the temporal arch (barbouri, and nigrinoda; polymorphic in geographica, G. oculifera, and texana; Fig. 24). For now, we scored all taxa as present, regardless of whether the contact is visible in lateral view, medial view, or both. In several other emydid taxa, the bones closely approach one another, but do not actually meet, on the inside of the temporal arch (alabamensis, ernsti, flavimaculata, gibbonsi, kohnii, P. nelsoni). Polarity: A contact is present between the quadratojugal and maxilla in odoratus, but absent in spinifera, caretta, serpentina, and Mongolemys. We consider the presence of a contact to be derived for Testudinoids. (13) Medial contact of the maxillae along the anterior margin of the jaw; 0 = absent, Figs. 25, 26; 1 = present, Fig. 27 (modified from Hirayama, 1985, 20; McCord et al., 1995, 2; Yasukawa et al., 2001, 10). In most testudinoids, the anteromedial ends of the maxillae are separated medially by the premaxillae along the anterior margin of the jaw (Fig. 25). Hirayama (1985) noted that the maxillae have a medial contact in some batagurids, which was confirmed by McCord et al. (1995) and Yasukawa et al. (2001) for spengleri and several other species that they included in the genus Geoemyda. We found a broad medial contact of the maxillae in spengleri and annulata (Fig. 27). In some species, the maxillae approach one another along the ventral rim of the nasal opening (e.g., amboinensis, mouhotii, pulcherrima, crassicollis), but a well-developed contact is never present (Fig. 26). Polarity: The maxillae do not meet medially along the anterior margin of the jaw in odoratus, caretta, serpentina, and Mongolemys. A medial contact is present in spinifera but only along the ventral border of the external nares. We consider a medial contact along the anterior margin of the jaw to be the derived condition within Testudinoidea. (14) Size of the foramen orbito-nasale; 0 = small, less than 1/6 of orbit length, Figs. 28, 29; 1 = large, more than 1/6 of orbit length, Fig. 30 (modified from Hirayama, 1985, 33; Gaffney and Meylan, 1988, F9.3, F10.2, G10.3, H11.1, H16.3; Crumly, 1982, 25; Crumly, 1994, 12). We were cautious when first approaching this character due to the inconsistent usage and definition of

8 Vol. 10, p. 60 Asiatic Herpetological Research 2004 small and large by various authors. However, after assessing the size of this foramen based on its size relative to the length of the orbit, we were surprised to see that we were able to reproduce Hirayama s (1985) scoring for the batagurids without too many difficulties. In contrast, our initial observations of tortoises were in stark contrast to those of Crumly (1982, 1985, 1994) and Gaffney and Meylan (1988). This may be due to the thin nature of the palatine of many tortoises, and the relative ease with which that part of the palate can be damaged during skeletal preparation and handling. Furthermore, the foramen becomes progressively more closed with increased ontogenetic age (Crumly, 1982). We encountered similar problems in attempting to score reticularia and blandingii. Because we deem this character to be potentially useful for helping to resolve phylogeny within batagurids and emydids, we decided to score all testudinoids with delicate palatines (i.e., all tortoises, reticularia, and blandingii) as unknown. We acknowledge that our redefinition of the character is still subjective and somewhat problematic, but using this definition we were able to unambiguously score all the ingroup taxa we examined. Polarity: The foramen orbito-nasale is large in serpentina, odoratus, and spinifera, but small in caretta. We consider presence of a large foramen orbito-nasale to be the derived condition within testudinoids, because the foramen is small in Mongolemys. (15) Contact between maxilla and vomer; 0 = present, Fig. 31; 1 = absent, vomer separated from the maxilla by the premaxilla, Fig. 32 (Hirayama, 1985, 31; Crumly, 1982, 21; Yasukawa et al., 2001, 14). We generally agree with previous scorings for this character (Hirayama, 1985, Crumly, 1982, Yasukawa et al., 2001). We confirm the absence of a contact in amboinensis and pulcherrima, but our scoring differs slightly for those taxa that Hirayama (1985) coded as intermediate apomorphic, a character state that we interpret as polymorphism. Of those taxa that Hirayama (1985) and Yasukawa et al. (2001) scored as intermediate (flavomarginata, caspica, annulata), our sample size is too small to confirm whether both character states are present. We consequently follow these authors by scoring those taxa as polymorphic. Polarity: A contact between the maxilla and vomer is present in all outgroups. The loss of this contact is derived for testudinoids. (16) Size of the foramen palatinum posterius; 0 = large, Fig. 33; 1 = small, Fig. 34 (modified from Hirayama, 1985, 22; Gaffney and Meylan, 1988, F2.2, F6.1; McCord et al., 1995, 4; Yasukawa et al., 2001, 12). Our characters 16 and 17 were published originally by Hirayama (1985) as one character that combined two morphological features: the size of the foramen palatinum posterius (f.p.p.) and participation of the pterygoid in the margin of the f.p.p. Although four possible combinations of these features are logically possible, only two were originally included (participation present, f.p.p. large; participation absent, f.p.p. small). Gaffney and Meylan (1988) also used this character within emydids, but their character applied only to the exclusion of the pterygoid from the f.p.p. We decided to subdivide Hirayama s (1985) character into one character that describes the size of the f.p.p. and a second that addresses the position of the pterygoid relative to the f.p.p. We found no difficulty in identifying the f.p.p. as large or small, (Figs. 33 and 34, respectively), and no ambiguous condition was encountered. Because our character definition only includes two character states, our scorings do not reflect those of other workers with a more limited target group (e.g., McCord et al., 1995). In juveniles the f.p.p. tends to be larger, but during later ontogenetic stages the f.p.p. is slowly reduced in size. Polarity: The f.p.p. is small in odoratus, spinifera, and most pleurodires; it is absent in caretta, but is large in serpentina and Mongolemys. We consider a small f.p.p. to be the derived state. (17) Position of the pterygoid relative to foramen palatinum posterius (f.p.p.); 0 = pterygoid situated posterior to the f.p.p., Fig. 33; 1 = pterygoid situated posterior to the f.p.p., but sends a process anterior and lateral to the f.p.p., Fig. 34. Our survey of testudinoids indicated that reliable assessment of participation of the pterygoid in the f.p.p. may be difficult because many species show an ontogenetic change in configuration of this part of the palate. In juveniles, the f.p.p. typically includes the pterygoid in its posterior margin. During later ontogenetic stages the pterygoid is excluded. In spite of this, it appears that the relative position of the pterygoid tends to stay constant during ontogeny. Polarity: The anterior end of the pterygoid is situated posterior to the f.p.p. in odoratus, serpentina, most pleurodires, and Mongolemys, but is situated lateral to the f.p.p. in spinifera. We consider a posterior position to be primitive for Testudinoidea. (18) Epipterygoid participation in the trigeminal foramen; 0 = absent, Fig. 35; 1 = present, epipterygoid clearly separates the parietal and pterygoid in lateral view, Fig. 36. The anteroventral rim of the trigeminal foramen of most testudinoid turtles is formed by the parietal and pterygoid. The epipterygoid commonly comes close to the foramen, but does not form part of it. In spengleri

9 2004 Asiatic Herpetological Research Vol. 10, p. 61 and mouhotii, the epipterygoid consistently participates in the margin of the trigeminal foramen thus separating the parietal and pterygoid, at least in lateral view. Polarity: The epipterygoid forms part of the anteroventral rim of the trigeminal foramen in the majority of outgroups, with the exception of pleurodires that lack a definitive ossified epipterygoid (Gaffney, 1979). The condition is unclear for Mongolemys. However, within the ingroup we found this character only in the seemingly rather specialized turtles spengleri and mouhotii. We consequently consider its presence to be secondarily derived. (19) Vomerine foramen; 0 = absent, Fig. 37; 1 = present, Fig. 38 (Gaffney and Meylan, 1988, H4.1; Crumly, 1994, 15). The vomerine foramen (= anteromedial vomerine aperture of Crumly, 1982 and 1994 [in part]) is a small opening that pierces the vomer along the midline just posterior to the foramen praepalatinum (Bramble, 1971). The presence of a vomerine foramen was noted in agassizii and berlandieri by Bramble (1971), and was used by Gaffney and Meylan (1988) to unite various species currently placed in Gopherus as a clade. Its irregular presence in elegans, elongata, chilensis, and radiata was reported by Crumly (1982). Specimens in our sample enable us to confirm its presence in agassizii, berlandieri, and chilensis. Polarity: The vomerine foramen occurs in only a few testudinids and it is absent in all outgroups. Its presence is considered derived for Testudinoidea. (20) Development of the foramen praepalatinum as a canal (canalis praepalatinum) that is concealed by a bony secondary palate in ventral view; 0 = absent, Fig. 39; 1 = present, Fig. 40 (modified from Hirayama, 1985, 24). In most testudinoids, the foramen praepalatinum is a small opening that connects the nasal cavity with the roof of the oral cavity (Fig. 39). However, in a number of taxa with extensively developed secondary palates, the anterior nasal artery passes through an elongated canal that is concealed in ventral view by the bony secondary palate (e.g., baska, tentoria, petersi; Fig. 40). We refer to this structure as the canalis praepalatinum. Our scorings are fully consistent with those of Hirayama (1985). Polarity: The foramen praepalatinum is absent in spinifera (Gaffney, 1979) and caretta (Nick, 1912), and is developed as a true foramen that is exposed in ventral view in odoratus, serpentina, and Mongolemys. The development of a canalis praepalatinum is considered the derived condition within Testudinoidea. (21) Contact between pterygoid and basioccipital; 0 = present, Fig. 41; 1 = absent, Fig. 42 (modified from Gaffney and Meylan, 1988, F1.1, F10.3, H18.3; Crumly, 1994; Shaffer et al., 1997, 103). Two of the most often-cited characters that purportedly help to distinguish the Emydidae from the Bataguridae are the batagurine process and the contact between the pterygoid and the basioccipital (McDowell, 1964). Both traits are commonly combined as one character (e.g., Gaffney and Meylan, 1988) and even seem to have been confused with one another (e.g., Mlynarski, 1976; Shaffer et al., 1997). The batagurine process is a poorly-defined feature that, in McDowell s original usage (1964) appears to consist of a lateral process of the basioccipital that floors the recessus scalae tympani. Many testudinoid species (including non- batagurines ) have a laterally-projecting process of the basioccipital; it may or may not floor the recessus scalae tympani, but it is often obscured from view in articulated specimens. When disarticulated material is examined, a broader distribution of this feature (assuming it is interpreted as we have done above) across testudinoids is revealed. Within Testudinoidea, the pterygoid commonly sends a process posteriorly and contacts the basioccipital just lateral to the basisphenoid. This character appears to be absent in most emydids, but is present in terrapin and those species that are currently attributed to Graptemys (Gaffney and Meylan, 1988). We noticed a strong ontogenetic component to this character within emydids. The pterygoid is commonly rather short during younger ontogenetic stages, but finally reaches the basioccipital in later stages. For instance, among our specimens of terrapin, orbicularis, and texana, the pterygoid does not contact the basioccipital in younger individuals, but a clear contact is present in adults. We score such species as polymorphic, but note that ontogenetically influenced polymorphisms are not well explored in turtles. Among batagurids, we noted a similar pattern. Contrary to general belief we were not able to observe a contact between the pterygoid and the basioccipital in all taxa traditionally classified in this group (e.g., trijuga, pulcherrima, sinensis). As with the emydids, there seems to be an ontogenetic effect in which younger specimens tend not to have a contact. Where recognized, we scored these species as polymorphic. We found much conflicting data regarding the distribution of this character among tortoises (Crumly, 1982, 1985, 1994; Gaffney and Meylan, 1988). Among the specimens we examined, we note the complete absence of a contact only in graeca; polymorphisms were observed in polyphemus and horsfieldi. Again, an ontogenetic component is apparent. Polarity: A contact between the pterygoid and

10 Vol. 10, p. 62 Asiatic Herpetological Research 2004 basioccipital is present in all outgroups. Its absence is considered to be derived. (22) Contact of the pterygoid with the articular facet of the quadrate; 0 = absent, Fig. 43; 1 = present, Fig. 44 (Hirayama, 1985, 38). According to the data matrix published by Hirayama (1985) he only observed this contact in reevesii, however, in his tree (fig. 2) the contact is mapped as an autapomorphy of subtrijuga. We confirm the presence of a contact between the posterior process of the pterygoid and the articular surface of the quadrate in subtrijuga. It is the only taxon we examined that displays the derived condition. Polarity: A contact between the pterygoid and the articular surface of the quadrate is absent in all outgroups and the vast majority of the ingroup. We consider its presence to be derived. (23) Closure and depth of the incisura columella auris; 0 = absent, incisura is open, Fig. 44; 1 = present, incisura closed, Fig. 45 (Crumly, 1985; Gaffney and Meylan, 1988, H1.3). The incisura columella auris is a notch that is formed by the quadrate and that holds the stapes and eustachian tube (Gaffney, 1972). In a number of turtles, the incisura closes to fully surround the stapedial shaft (Gaffney and Meylan, 1988). Within testudinoids, the incisura evidently is closed in most tortoises (Crumly, 1985). We are able to confirm the presence of such a closed incisura in all testudinids we examined with the exception of one specimen of kleinmanni (CAS ), the smallest of the species now classified in Testudo. In a number of batagurids and emydids the incisura commonly is very narrow and even appears to be closed, however, a closer look under the microscope combined with a probing needle reveals that this apparent closure is produced by dry tissues that remain in this area in many articulated skulls. The only batagurid for which we sometimes found a closed incisura is N. platynota; in that species the quadrate does not fuse together forming a solid ring behind the incisura, but this is also the case for many tortoises (e.g., some belliana, emys, some homeana, some kleinmanni). In some cases, the polymorphism we noted (e.g., in belliana and homeana) appears to be a result of ontogenetic age, with older individuals displaying a greater degree of fusion at the posterior part of the incisura. Polarity: The polarity of this character is somewhat ambiguous, because the incisura columella auris is closed in serpentina and spinifera, but open in caretta and odoratus. We conclude that its presence is derived within testudinoids because it is absent in Mongolemys. Mandible (24) Angular contribution to the sulcus cartilaginis Meckelii; 0 = present, the angular contributes to the sulcus and is as long or longer than the prearticular, Fig. 47; 1 = absent, the angular is shorter than the prearticular, Fig. 48 (modified from Gaffney and Meylan, 1988, F1.4). A broad contact of the angular with Meckel s cartilage was used by McDowell (1964) to characterize the Emydinae and later used by Gaffney and Meylan (1988) as a synapomorphy to unite the same grouping. As it was originally worded, this character is difficult to observe in its literal sense for most museum specimens, because the Meckel s cartilage usually is not present in modern and fossil skeletal specimens. We suggest that the spirit of McDowell s (1964) character can be evaluated by examining the participation of the angular in the sulcus cartilaginis Meckelii. We confirm that a small to broad angular contribution is present in all emydids with the exception of rubriventris. In most batagurids, the angular is a short bone that does not participate in the sulcus and barely spans half the distance the prearticular does. However, a small but clear contribution to the sulcus is present in an eclectic group comprised of baska, dentata, thurjii, punctularia, and some grandis. We were not able to carefully evaluate potential polymorphisms in these taxa. Polarity: An angular contribution to the sulcus cartilaginis Meckelii is present in all cryptodiran outgroup taxa. We consider is absence to be derived. (25) Contact between surangular and dentary; 0 =simple contact, Fig. 49; 1 = strongly interdigitated suture, Fig. 50 (Crumly, 1982, 12; Crumly, 1985; Gaffney and Meylan, 1988, H6.1). In most testudinoids, the surangular and dentary meet along the lateral side of the mandible in a simple, overlapping contact. According to Crumly (1982, 1985), this contact is stabilized through a finger-like process of the surangular that interdigitates with the dentary in all tortoises except emys, impressa, and those species he classified in Gopherus. We confirm the absence of this character in agassizii, berlandieri, emys, impressa, and polyphemus, but also did not observe it in areolatus. Polarity: An interdigitated contact between the surangular and dentary is absent in all outgroups and the majority of the ingroup. We consider its presence to be derived. (26) Height of the processus coronoideus; 0 = as high as dentary, Fig. 51; 1 = rising significantly above the dentary, Fig. 52 (modified from Hirayama, 1985, 45). The coronoid of most turtles is a small bone that produces a minor knobby projection that rises only little

11 2004 Asiatic Herpetological Research Vol. 10, p. 63 above the adjacent dentary, if at all. According to McDowell (1964) and Hirayama (1985) the coronoid is larger and rises moderately above the dentary in borneensis and crassicollis, and in reevesii and subtrijuga the coronoid is very large and produces a robust process that sits high above the dentary. We confirm these observations, however, we were also able to observe moderately developed coronoid processes in kachuga and tentoria. Among emydids, we also observed moderately developed coronoids in barbouri, ernsti, flavimaculata, geographica, gibbonsi, kohnii, and terrapin. Unlike Hirayama (1985), we only utilize one derived character state, because it is difficult to objectively measure and discretize the relative height of the coronoid among turtles. Polarity: The coronoids of caretta, serpentina, and odoratus are small and do not rise above the dentary, but the coronoid of spinifera is well developed and forms a moderate process. The lower jaw is not described for Mongolemys. We consider well-developed coronoids to be derived within testudinoids. (27) Foramen dentofaciale majus; 0 = small, Fig. 53; 1 = large and situated within a large lateral fossa, Fig. 54 (Hirayama, 1985, 47). The foramen dentofaciale majus of most testudinoids is a small opening that is situated on the lateral side of the mandible, just ventral and slightly anterior to the coronoid. The foramen dentofaciale majus is greatly enlarged in thurjii and ocellata and is situated at the anterior end of an expanded lateral fossa (Hirayama. 1985). We confirm the enlargement in those taxa, and further note its presence in petersi. Polarity: The foramen dentofaciale majus is small in all outgroups and the vast majority of the ingroup. Its presence is considered to be derived. Triturating Surfaces (28) Participation of palatine in the triturating surface of the upper jaw; 0 = absent, Fig. 55; 1 = present, Figs. 56, 57 (Hirayama, 1985, 26; Gaffney and Meylan, 1988, F2.1). In some testudinoids, the palatine has a ventrolateral maxillary process that participates in the triturating surface of the upper jaw. The degree of participation varies among taxa, and within some species. A clear and extensive participation is present in barbouri, ernsti, geographica, gibbonsi, petersi, rubriventris, scripta, terrapin, texana, and versa. It is weakly developed in hamiltonii, ocellata, G. oculifera, ouachitensis, reevesii, subtrijuga, and some individuals of baska. Participation was used by Gaffney and Meylan (1988) to unite Terrapene spp., blandingii, guttata, G. insculpta, marmorata, muhlenbergii, and orbicularis as a clade within the Emydidae. In our observations, however, this participation also is absent among other emydids such as flavimaculata, kohnii, nigrinoda, picta, and reticularia (we were not able to evaluate adequately the potential for polymorphism in these taxa). In addition, it appears that the absence of a palatine participation represents the plesiomorphic state for testudinoids, thus eliminating its value for diagnosing monophyletic groups. Polarity: The palatine does not participate in the triturating surface of spinifera, but does so in odoratus, caretta, and serpentina. The palatine does not participate in the triturating surface of Mongolemys and only occurs in Testudinoids with highly derived secondary palates. We consequently consider the participation of the palatine in the triturating surface to be derived within testudinoids. (29) Participation of the vomer in the triturating surface of the upper jaw; 0 = absent, Figs. 56, 57, 58; 1 = present, Fig. 59 (Hirayama, 1985, 25). The triturating surface is the grinding surface of the jaw. In most turtles, it is formed on the upper jaw predominantly by the maxilla and premaxilla. However, in turtles with extensive secondary palates the vomer may also participate. In texana, the vomer may have a ventral projection that barely separates the maxillae in the midline, but it does not participate in the triturating surface proper because it sits in a dorsal concavity within the palate (Fig. 56). Our scorings differ from those of Hirayama (1985) for mouhotii (which Hirayama scored as intermediate apomorphic and we score as absent because it does not have a secondary palate) and subtrijuga (in our specimens the vomer does not descend to the palatal surface, but this species may be polymorphic). This character is polymorphic in barbouri. Polarity: Because the vomer does not participate in the triturating surface of odoratus, spinifera, caretta, serpentina, and Mongolemys, its participation is considered to be the derived condition for Testudinoidea. (30) Presence and number of lingual ridges of the triturating surfaces of the upper and lower jaws; 0 = no lingual ridges present, Fig. 60; 1 = one lingual ridge present, Figs ; 2 = two lingual ridges present (modified from Hirayama, 1985, 29, 44; Gaffney and Meylan, 1988, F7.2, F9.1) Most turtles lack lingual ridges on their triturating surfaces (Fig. 60), but one or two such ridges are developed in a number of testudinoids (Figs ). These ridges run parallel to the labial surface of the maxilla and dentary, and typically do not meet their counterpart on the midline. They are not necessarily a continuous structure (Fig. 61), and may be divided into several com-

12 Vol. 10, p. 64 Asiatic Herpetological Research 2004 ponents. In some cases, an extensive ridge-like structure can create the appearance of an additional ridge at the extreme lingual margin of the maxilla bordering the internal nares; however, we consider these to be the thickened rim of the internal nares rather than an additional ridge. Among batagurids and emydids, one lingual ridge is present in alabamensis, borneoensis, decorata, kachuga, P. nelsoni, ocellata, petersi, rubriventris, scripta, sinensis, tentoria, texana, and thurjii. Two lingual ridges are developed in baska. We found lingual ridges in all tortoises we examined except erosa, belliana, and homeana. Hirayama (1985) originally scored this character as two separate characters, one for the mandible and one for the maxilla. In our observations, the triturating surface of the lower jaw closely mimics that of the upper jaw, creating an occlusal surface that closely reproduces the function of cusps in mammalian cheek teeth. Both characters were scored in parallel in Hirayama s matrix, and we see no reason to consider them independent. Polarity: All outgroups lack lingual ridges on the triturating surfaces. We consequently consider their presence to be derived. (31) Well-developed serrations on labial or lingual ridges of the triturating surfaces of the upper and lower jaws; 0 = absent, Fig. 60; 1 = present, Fig. 61 (modified from Hirayama, 1985, 21, 27, 41, 43, 46; Gaffney and Meylan, 1988, F9.2; Yasukawa et al., 2001, 11). Well-developed serrations on the lingual and labial ridges of the upper and lower jaws are developed in several batagurids and emydids. A number of tortoises and other batagurids (e.g., carbonaria, pardalis, sulcata, annulata, and areolata) exhibit serrations on their ramphothecae, but unlike the bony, tooth-like serrations of borneoensis, thurjii, petersi, or texana, these crenulations are weakly developed, leaving very little or no trace of serrations on the underlying bone. In comparison to those taxa with well-developed serrations, it very difficult to establish a consistent scoring system for taxa with fine crenulations, because many specimens will not exhibit any serrations, probably due to wear of the ramphothecae. Unlike Hirayama (1985) and Yasukawa et al. (2001) we scored all taxa with such weak serrations as absent. Unfortunately, even in those taxa with well-developed serrations, the serrations are not always evenly developed on all ridges. We consequently combined all of Hirayama s (1985) characters relating to serrations into one character. Because serrations commonly occur on all available ridges, this treatment will also help to avoid unconsciously weighting the presence of serrations with up to five characters. In those taxa that have them, the ridges themselves often have very different morphologies; this character needs to be critically reevaluated with adequate sample sizes for the relevant taxa. Polarity: All of our outgroup taxa and the majority of the ingroup taxa lack strong serrations. We interpret their presence to be derived. (32) Median ridge or sulcus of the triturating surface of the upper jaw; 0 = both structures absent, Fig. 60; 1 = median ridge present, Fig. 62; 2 = median sulcus present, Fig. 63 (modified from Hirayama, 1985, 30; Crumly, 1985, 1994, 4; Gaffney and Meylan, 1988, H3.1). In a number of testudinoid turtles with partially developed secondary palates and lingual ridges, additional structures are formed along the midline of the upper jaw that typically correspond to reciprocal structures of the lower jaw. The upper jaw of petersi is characterized by a narrow sulcus (Fig. 63) and the mandible exhibits a low median ridge. On the other hand, in baska, borneoensis, thurjii, kachuga, agassizii, berlandieri, and polyphemus, a ridge runs along the midline (Fig. 62), which typically corresponds to a sulcus in the lower jaw. An incipient ridge also was reported in emys (Crumly, 1994), but we were not able to confirm this on the specimen available to us. Polarity: A median ridge is absent in all outgroups and the vast majority of the ingroup. Its presence is considered to be derived. (33) Posterior extension of the lower triturating surface behind the symphysis of the dentary; 0 = absent, Fig. 64; 1 = present, Fig. 65 (Hirayama, 1985, 42; Gaffney and Meylan, 1988, G5.2). In several batagurids, the triturating surface of the dentary forms a shelf along the midline that extends so far posteriorly that the symphysis cannot be seen when the mandible is observed in dorsal view (McDowell, 1964). Our scorings fully agree with those of Hirayama (1985) for the Bataguridae, but we disagree with Gaffney and Meylan (1988) who asserted that this character also occurs in some Emydidae. Admittedly, several species currently placed in the genera Graptemys, Pseudemys, and Trachemys have greatly expanded triturating surfaces of the dentary, but in all of the specimens available to us, the symphysis is always visible in dorsal view. Polarity: An extended triturating surface of the dentary does not occur in any outgroup taxon. We consider its presence to be derived. Carapace (34) Carapace strongly tricarinate in adult; 0 = absent, Figs. 66, 67; 1 = present, Fig. 68 (modified from

13 2004 Asiatic Herpetological Research Vol. 10, p. 65 Hirayama, 1985, F; McCord et al., 1995, 10; Yasukawa et al., 2001, 24). Three distinct carapacial ridges are present in the adults of reevesii, hamiltonii, spengleri, subtrijuga, trijuga, and mouhotii. We cannot replicate Hirayama s (1985) placement of this character as an autapomorphy in hamiltonii. In our observations, the carinae in hamiltonii are not better developed than in some other taxa. In fact, they are more weakly developed than those in mouhotii (Fig. 68). Because keels are present in the young and subadults of such taxa as crassicollis (Fig. 67), mutica, and sinensis, but disappear with age, and because we were not able to observe the juveniles of most species, we restricted this character to those species that exhibit well-developed tricarinae as adults. Three keels were reported to be present in the adults of dentata (McCord et al., 1995), but we cannot confirm this (tricarinae are not present on our younger specimens). Polarity: Tricarinae are absent in our outgroup species (caretta, odoratus, serpentina, spinifera, gibba, siebenrocki, subglobosa, and subrufa), but do appear occasionally in some of their close relatives, such as scorpioides, temminckii, and fimbriatus. We consider the presence of tricarinae to be derived within Testudinoidea. (35) Significant serration of the posterior peripherals; 0 = absent, Fig. 66; 1 = present, Fig. 68 (modified from Hirayama, 1985, D; McCord et al., 1995, 11; Yasukawa et al., 2001, 23). We generally agree with previous observations reported for this character (Hirayama, 1985). However, because the carapace rim is at least slightly serrated in almost all turtles, we rephrase the character definition to include only significantly serrated posterior peripherals as found, for example, in crassicollis, dentata, grandis, mouhotii, N. platynota, and spengleri. Among the emydids and testudinids, the peripherals of barbouri, erosa, flavimaculata, geographica, homeana, kohnii, nigrinoda, oculifer, G. oculifera, pseudogeographica, and versa also are serrated. It is important to note that our scores are based on the peripheral bones; the amount of carapacial serration greatly depends on the presence or absence of the marginal scutes in the specimens used, because the scutes greatly accentuate the amount of serration, if present. We find no conflict with the codings of McCord et al. (1995) and Yasukawa et al. (2001). Polarity: Serrated posterior peripherals are present in caretta, but absent in odoratus, serpentina, and most pleurodires. However, due to ingroup commonality and its absence in taxa placed within Lindholmemydidae we conclude that its presence is derived. (36) Carapace of adult tectiform in cross-section with a strong posterior projection on the third vertebral scute; 0 = absent, Fig. 69; 1 = present, Fig. 70 (Hirayama, 1985, N). According to Hirayama (1985), this character only occurs in tecta and tentoria. For our sample, we were able to confirm its presence in tecta and tentoria and also observed it in barbouri (barbouri and other species now classified in Graptemys may be sexually dimorphic for this character). The descriptive term tectiform is somewhat problematic, because any turtle shell can be considered roofed (Fig. 69). We regard a carapace as tectiform if its sides are more-or-less flat and meet along the midline at a rather sharp angle (Fig. 70). Many testudinoids, and notably those emydids currently classified within Graptemys, have a somewhat tectiform carapace as juveniles, but that morphology typically is lost in the adults. Polarity: Because all outgroups and the majority of the ingroup do not have a tectiform carapace, we consider its presence to be derived. (37) Shape and orientation of the second neural; 0 = second neural hexagonal, short sides positioned anteriorly, Fig. 71; 1 = second neural hexagonal, short sides positioned posteriorly, Fig. 72; 2 = second neural octagonal, Fig. 73 (modified from Hirayama, 1985, G; Yasukawa et al., 2001, 25). (38) Shape and orientation of the third neural; 0 = third neural hexagonal, short sides positioned anteriorly, Fig. 71; 1 = third neural hexagonal, short sides positioned posteriorly, Fig. 72; 2 = third neural square, Fig. 74; 3 = third neural octagonal, Fig. 75 (modified from Hirayama, 1985, G; Yasukawa et al., 2001, 25). Originally, Hirayama (1985) only discussed the orientation of the neurals in general, which is fully sufficient for his batagurid ingroup, because almost all individuals exhibit his two suggested character states. However, in most tortoises the second and/or third neurals are not hexagonal, but rather are square or octagonal, making it impossible to assign them to one of Hirayama s (1985) character states. We consequently split Hirayama s original character into two characters, restricted their application to the second and third neural, and added additional character states. Our observations generally agree with those of Hirayama (1985) and Yasukawa et al. (2001) for batagurids, with the exception of annandalei in which we found the short side of the second and third neurals to be positioned anteriorly, and not posteriorly as was indicated by Hirayama (1985). Polarity: The short side of the second and third neural bones faces posteriorly in odoratus and spinifera but

14 Vol. 10, p. 66 Asiatic Herpetological Research 2004 anteriorly in caretta. The shape of the second and third neurals is extremely variable in serpentina. However, in Lindholmemys and Mongolemys the short side of the second and third neurals is positioned anteriorly. We consider that condition to be primitive for Testudinoidea. (39) Medial contact of the seventh and/or eighth costal bones; 0 = absent, Fig. 76; 1 = present, Fig. 77 (Hirayama, 1985, V; Yasukawa et al., 2001, 26). In some testudinoid turtles, the posterior costal bones meet along the midline due to the reduction of the posterior neural elements. The original character definition provided by Hirayama (1985) and Yasukawa et al. (2001) was worded to indicate a contact between the seventh and eighth costal bones among some batagurids. We are unable to reproduce their results if the character definition is taken literally. For example, in all our specimens of amboinensis and galbinifrons, the eighth costals meet on the midline, but the seventh costals do not. However, if the character definition is modified to include any contact of the seventh or eighth costals, our results are concordant with those of Hirayama (1985) and Yasukawa et al. (2001). In addition to the Asian box turtles, we report a medial contact of the posterior costals in baska, carolina, coahuila, T. nelsoni, T. ornata and rubida. Polarity: A medial contact of the seventh and/or eighth costals is absent in serpentina, but present in spinifera, caretta, many kinosternids, and many pleurodires. Although a composite reconstruction of a putative lindholmemydid from the Early Cretaceous of Japan was illustrated with the seventh costals in contact at the midline (Hirayama et al., 2000, fig. 11), such a contact is absent in other specimens of Lindholmemys and Mongolemys. Its presence within testudinoids is predominantly in the highly derived box turtles, and we conclude that its presence is derived for Testudinoidea. (40) Cervical scute; 0 = present, Fig. 78; 1 = absent, Fig. 79 (modified from Crumly, 1985, 1994, 34; Gaffney and Meylan, 1988, H5.2, H10.1; Shaffer et al., 1997, 41). The presence and shape of the cervical scute is used commonly to determine phylogenetic relationships within tortoises. According to Crumly (1985) the cervical scute is very narrow or absent in all tortoises except agassizii, berlandieri, emys, flavomarginatus, impressa, and polyphemus. We generally agree with these observations, but when this character is applied to all testudinoids intraspecific variation is so great that the character becomes essentially useless. We consequently limit our scoring to the mere presence or absence of the cervical scute. We confirm the observations of Gaffney and Meylan (1988) that this scute is absent in carbonaria, chilensis, elegans, nigra, pardalis, and sulcata and additionally code homeana and erosa as polymorphic. Polarity: The cervical scute is present in all cryptodiran outgroups that have scutes on their carapace. We consider its absence to be derived. (41) Number of vertebral scutes; 0 = five, Fig. 80; 1 = six or more, Fig. 81 (Hirayama, 1985, P). We confirm Hirayama s (1985) observation that there are at least six vertebral scutes in N. platynota. Additional scutes occasionally occur in other species, but are best considered abnormalities; they typically lack the symmetrical associations with adjacent pleural scutes seen in N. platynota. Polarity: All testudinoids, except for N. platynota, have five vertebral scutes. We consider the presence of six scutes to be derived. (42) Position of the anterior sulcus of the fourth vertebral scute; 0 = sulcus lies on the fifth neural, Fig. 82; 1 = sulcus lies on fourth neural, or on the suture between the fourth and fifth neural, Fig. 83; 2 = sulcus lies on the sixth neural, or on the suture between the fifth and sixth neural, Fig. 84 (modified from Hirayama, 1985, L+M). (43) Position of the posterior sulcus of the fourth vertebral scute; 0 = sulcus lies on the eighth neural, or on the homologue of the eighth neural, if the seventh is reduced (e.g., in most tortoises), Fig. 85; 1 = sulcus lies on the seventh neural, or on the suture between the seventh and eight neural, Fig. 86; 2= eighth neural absent, sulcus overlies costals that meet at the midline, Fig. 87 (modified from Hirayama, 1985, L+M). The size of the fourth vertebral scute was addressed with two characters by Hirayama (1985) but the total range of morphological variability in testudinoids is not encompassed by his character definitions. In most testudinoids, the fourth vertebral scute covers the posterior half of the fifth neural bone, the sixth and seventh neurals, and the anterior half of the eighth neural. A number of variations are known, and simply counting the number of neural bones covered by this scute results in problems by creating a false perception of homology. For instance, a fourth scute that partially overlies the fourth and seventh neurals and fully covers the fifth and sixth, can strictly be said also to cover four neurals, but the elements involved are only partially homologous with the common condition. We attempt to resolve these issues by establishing two new characters that preserve what we think was Hirayama s (1985) original intent, but that permit a more accurate representation of the association between the fourth vertebral scute and the underlying neural bones.

15 2004 Asiatic Herpetological Research Vol. 10, p. 67 Some problems that are associated with scoring this character include the prevalence of scute abnormalities among testudinoids (e.g., Coker, 1905, 1910; Newman, 1906; Zangerl and Johnson, 1957). Specimens exhibiting such abnormalities were scored as unknown. The notable exception to this is N. platynota, in which a sixth, or even a seventh, vertebral scute is always present. Polarity: Determining the polarity through outgroup relationship is somewhat difficult, because almost every outgroup exhibits a different condition, especially for the posterior sulcus. Based on ingroup commonality, and the presence of our zero state in both Lindholmemys and Mongolemys, we consider the sulci of the fourth vertebral scute to be primitively situated on the fifth and eighth neural bones. (44) Posterior margin of first vertebral scute significantly narrower than its anterior margin; 0 = absent, Fig. 88; 1 = present, Fig. 89 (modified from Hirayama, 1985, C). When originally proposed, this character was applied to a posterior constriction of all vertebral scutes (Hirayama, 1985). If strictly applied, this character is absent in all taxa, because the fifth vertebral scute never is constricted along its posterior edge relative to the anterior edge. If each scute is viewed by itself, it becomes apparent that especially the fourth vertebral scute tends to be narrowed posteriorly, as can be observed in all species now classified in the genera Graptemys, Heosemys, Trachemys, and Testudo among others. According to Hirayama (1985), posterior narrowing is limited to crassicollis and borneensis and unites them as a synapomorphy. We were able to replicate this distribution only if the character definition was restricted to the first vertebral scute, in which the posterior margin is significantly narrower than its anterior margin in those two species only. In making this change, however, this character becomes at least partly redundant with characters 45 and 47. Polarity: Outgroup analysis reveals that posterior narrowing of the first vertebral scute is present only in odoratus; it is absent in Lindholmemys and Mongolemys. We regard the presence of a posterior narrowing of the first vertebral scute to be derived. (45) Anterior half of the first vertebral scute much narrower than posterior half, especially in adults; 0 = absent, Fig. 90; 1 = present, Fig. 91 (modified from Hirayama, 1985, R). We confirm the clear presence of an anteriorly narrowed first vertebral scute in dentata and spinosa as reported by Hirayama (1985), and note that grandis is polymorphic. Because the anterior sulcus of the first vertebral scute commonly is restricted to the nuchal bone in several other taxa, but the scute shows no anterior narrowing, we limit the original character definition to the shape of the first vertebral scute only. This peculiar morphology seems to be the result of growth that is limited to the anterior edge and the posterior half of the lateral edge of the first vertebral scute. As a consequence, this character is not apparent in juveniles, but becomes increasingly accentuated in adults. Polarity: Anterior narrowing of the first vertebral scute is absent in all outgroups and within the large majority of the ingroup. We consider its presence to be derived. (46) Significant contact of the tenth marginal scute with the fifth vertebral scute; 0 = absent, Fig. 92; 1 = present, Fig. 93 (modified from Hirayama, 1985, K). Contact of the tenth marginal scute with the fifth vertebral scute was reported previously only in baska, smithii, tecta, and tentoria (Hirayama, 1985). We are able to confirm the presence of a very well developed contact in all but smithii (not seen), and we add spinosa to the list of species in which this contact may occur (it is polymorphic for spinosa; contact is present in CAS , but absent in the smaller CAS , so ontogenetic differences may explain the polymorphism). We also note slight contacts in some specimens of other species (e.g., agassizii, borneoensis, carbonaria, homeana, orbicularis, pardalis, and polyphemus), but by rewording Hirayama s (1985) original character to include only significant contact, we are able to retain what we believe was his original intent. Polarity: Due to the absence of contact in all outgroups in which it is applicable and the predominance within the ingroup, we consider its presence to be derived. (47) Contact of the second marginal scute with the first vertebral scute; 0 = absent, Fig. 94; 1 = present, Fig. 95 (Hirayama, 1985, O; see also Tinkle, 1962, table 1, Seam A ). According to Hirayama (1985), the first vertebral scute usually (>90%) contacts the second marginal scute in japonica, leprosa, and caspica. He also noted that the scutes are sometimes in contact in N. platynota and bealei. For our sample, we are able to confirm this contact as a polymorphism for bealei, caspica, japonica, and leprosa, but the contact is clearly absent in all our specimens of N. platynota. We also note that these scutes are sometimes in contact in picta, amboinensis, orbicularis, and terrapin. Together with all of the above, these taxa were scored as polymorphic. The only taxa to exhibit a consistently well-developed contact are reticularia and blandingii.

16 Vol. 10, p. 68 Asiatic Herpetological Research 2004 Polarity: A clear contact between the second marginal scute and the first vertebral scute does not exist in caretta or serpentina, but both morphologies occur in kinosternids and pleurodires. The scutes are not in contact in Lindholmemys and Gravemys, but they are in contact in Mongolemys. The polarity for this character is ambiguous. (48) Contact of the sixth marginal scute with the third pleural scute; 0 = absent, Fig. 96; 1 = present, Fig. 97 (modified from Hirayama, 1985, B; see also Tinkle, 1962, table 3, Seam C ). The contact between the sixth marginal and third pleural scutes is easily enough rendered as a simple presence or absence character, but this hides the range of possible morphological variation. The degree of contact can range from extensive to a condition where the two scutes just barely contact at their corners. Several taxa exhibit a condition where these scutes either barely touch or do not touch one another at their corners, but whenever several specimens were available to us, they typically turned out to be polymorphic. For this reason Gaffney and Meylan (1988) called this character dubious. We scored such borderline cases as polymorphic, even if not enough specimens were available to corroborate this. We confirm Hirayama s (1985) observations regarding the absence of a contact between these scutes in baska, ocellata, tecta, tentoria, and thurjii, and we add petersi and spinosa to that list. Taxa that we score as polymorphic include annandalei, borneensis, borneoensis, caspica, crassicollis, grandis, hamiltonii, japonica, punctularia, reevesii, sinensis, subtrijuga, and trijuga. Whereas all emydids lack a contact, tortoises exhibit both character states. The presence of a contact between the sixth marginal scute and the third pleural scute was considered by Hirayama (1985) to unite crown group batagurids and testudinids as a synapomorphy. Given the patchy distribution of this character, and widespread polymorphism, it seems to be of little use. Polarity: A contact between the sixth marginal scute and the third pleural scute is absent in most outgroups. The exception is caretta; this is not surprising because caretta has five instead of four pleural scutes. We consider the presence of a contact to be derived. (49) Twelfth marginal scute; 0 = two present, their common sulcus only partially subdivides the pygal bone, Fig. 98; 1 = two present, but their common sulcus fully subdivides the pygal bone, Fig. 98; 2 = both twelfth marginal scutes fused along the midline, Fig. 99 (modified from Mlynarski, 1976; Crumly, 1985, 1994, 35; Gaffney and Meylan, 1988, H2.1). According to McDowell (1964, p. 240 and table 1 number 4) members of the Emydinae can be distinguished from the Batagurinae based on an incomplete subdivision of the pygal bone by the median sulcus of the posterior-most marginals. We found exceptions with picta, N. platynota, pulcherrima, reevesii, reticularia, and spengleri, which do not always clearly exhibit the pattern that would be predicted by McDowell s (1964) statement, but we note that the expression of this character will depend significantly on the shape of the pygal bone. In all tortoises except emys and impressa, the twelfth marginal scutes are fused to from a single supracaudal scute that covers the posterior part of the carapace (Crumly, 1985). For this condition, we created a third character state. Polarity: The twelfth marginal scutes are fully separated in all outgroups that have them, and their common sulcus fully subdivides the pygal bone. The pygal bone in lindholmemydids is polymorphic, with the 0 state found in Lindholmemys and Mongolemys, and the 1 state in Gravemys. Either state 0 or state 1 is primitive for Testudinoidea; the midline fusion of the twelfth marginals is a derived feature for Testudinidae. Bridge (50) Sutured contact between plastron and carapace; 0 = present, plastron and carapace are tightly connected by an osseous bridge, Fig. 100; 1 = absent, plastron is attached to carapace by connective tissue, Fig. 101 (modified from Hirayama, 1985, Q; Shaffer et al., 1997, 58; Yasukawa et al. 2001, 21a). (51) Presence and development of anterior buttresses; 0 = anterior buttresses absent, Fig. 102; 1 = anterior buttresses present but small, and not in contact with the first costal bones, Fig. 103; 2 = anterior buttresses long and thin and just barely in contact with the costal bones, if at all, Fig. 104; 3 = anterior buttresses well developed and in clear contact with the first costal bones, Fig. 105; 4 = anterior buttresses very large and in direct contact with the first dorsal rib, Fig. 106 (modified from Hirayama, 1985, Q; Gaffney and Meylan, 1988, A14.2; Yasukawa et al., 2001, 28). (52) Presence and development of posterior buttresses; 0 = posterior buttresses absent, Fig. 107; 1 = posterior buttresses present but small, and not in contact with the costal bones, Fig. 108; 2 = posterior buttresses long and thin and just barely in contact with the costal bones, if at all, Fig. 109; 3 =posterior buttresses well developed and in clear contact with costal bones V and VI, Fig. 110; 4 = posterior buttresses well developed but only in clear contact with costal bone V, Fig. 111 (modified from Hirayama, 1985, Q; Gaffney and Meylan, 1988, A14.2, Shaffer et al., 1997, 55; Yasukawa et al., 2001, 29).

17 2004 Asiatic Herpetological Research Vol. 10, p. 69 (53) Medially-directed pivoting process for plastral hinge developed on fifth peripheral bone; 0 = absent, Fig. 112; 1 = present, Fig (54) Complete or almost complete overlap of hyoplastron/hypoplastron suture by the pectoral/abdominal sulcus; 0 = absent, Fig. 114; 1 = present, Fig. 115 (modified Gaffney and Meylan, 1988, F3.2; Burke et al., 1996, 16; McCord et al., 1995, 13; Yasukawa et al., 2001, 21b). In most testudinoid turtles, the plastron is attached to the carapace via a fully ossified bridge and variably developed plastral buttresses. In species with a kinetic plastron, the bridge is typically absent and the plastron is attached to the carapace via connective tissues (e.g., amboinensis, blandingii, carolina, dentata, galbinifrons, mouhotii, orbicularis, T. ornata). The original configuration of this character tied the presence of plastral kinesis to the reduction of the buttresses (Hirayama, 1985). However, within testudinoid turtles the morphology of the buttresses varies significantly and independently from plastral kinesis. We consequently split this character into three discrete characters concerned with the morphology of the bridge and the buttresses. We also developed two new characters that pertain to the morphology of the bridge region: the presence of medially-directed processes on the fifth peripherals that act as pivots for the plastral bones during shell closure (Bramble, 1974) and a revised plastral kinesis character (Gaffney and Meylan, 1988; McCord et al., 1995) that considers plastral-kinesis to be well developed only in those taxa in which the pectoral/abdominal sulcus fully overlaps the hyoplastron/hypoplastron suture, allowing optimal movement between the two plastral lobes. Most taxa with plastral kinesis also have well-developed pivoting processes on the fifth peripherals, but notable exceptions are orbicularis and N. platynota. In blandingii, the process is modified into an anteroposteriorly-elongated, flattened process that extends along most or all of the length of the fifth peripheral (Fig. 113). In some specimens of blandingii, a similar structure is developed on the sixth peripheral as well. Polarity: Reconstructing the basal condition for these characters within testudinoids is difficult, because all living cryptodiran outgroups do not have plastral buttresses and commonly lack osseous bridges. However, the bridge of Gravemys, Mongolemys, and Lindholmemys is osseous, shows no signs of kinesis, and (at least in Lindholmemys) the anterior and posterior buttresses are well developed and touch the costal bones. We consider that morphology to be primitive for Testudinoids. (55) Contact between inguinal and femoral scutes; 0 = absent, Fig. 116; 1 = present, Fig. 117 (Crumly, 1985, 1994, 42; Gaffney and Meylan, 1988, H3.3, H15.2). Within tortoises the complete or frequent absence of a contact between the inguinal scute and the femoral scute was used previously to hypothesize the monophyly of several smaller clades, for example {graeca + hermanni + horsfieldi + kleinmanni + marginata + tornieri} (Crumly, 1985) and {agassizii + berlandieri + flavomarginatus + polyphemus} (Gaffney and Meylan 1988). We confirm the absence of a contact in representatives of the first group, but not in the second. Among the second group (traditionally classified together in Gopherus) the contact is strongly reduced, but still is present. Among batagurids and emydids, a contact is absent in all taxa with the noteworthy exception of hamiltonii. Polarity: Determining the polarity for this character is somewhat difficult because all living outgroups have an arrangement of plastral scutes that is rather different from testudinoids. However, based on ingroup commonality and the absence of a contact in the Lindholmemydidae, we conclude that the presence of a contact between the inguinal and femoral scutes should be considered derived for Testudinoidea. (56) Presence of musk glands; 0 = inguinal and axillary gland present; 1 = axillary gland present only; 2 = musk glands absent (modified from Crumly, 1985). (57) Presence of anterior musk duct foramina; 0 = musk glands and their foramina present, Fig. 118; 1 = musk glands present, but foramina not developed; 2 = musk glands and foramina absent (modified from Hirayama, 1985, A; Gaffney and Meylan, 1988, F1.2, F5.3; Burke et al., 1996, 20). (58) Presence of posterior musk duct foramina; 0 = musk glands and their foramina present, Fig. 119; 1 = musk glands present, but foramina not developed; 2 = musk glands and foramina absent (modified from Hirayama, 1985, A; Gaffney and Meylan, 1988, F1.2, F5.3; Burke et al., 1996, 20). According to Hirayama (1985), the presence of musk duct foramina characterizes the paraphyletic assemblage Batagurinae (sensu Hirayama, 1985, not Gaffney and Meylan, 1988). We believe the difference of opinion between Hirayama (1985) and Gaffney and Meylan (1988) regarding this character is based on failure to make clear the distinction between the presence of musk glands and the presence of musk duct foramina. Musk glands are developed in almost all extant turtles (Waagen, 1972), and we consequently agree with Gaffney and Meylan (1988) that their presence should be considered primitive for all cryptodiran turtles.

18 Vol. 10, p. 70 Asiatic Herpetological Research 2004 However, even though most turtles have musk glands, true musk duct foramina are developed only in some pleurodires (e.g., Chelodina, Emydura), some emydids, and all batagurids, making a monophyletic Testudinoidea (sensu Hirayama, 1985) possible. Distinct musk duct grooves are present on the anterior peripherals of Kinosternidae, and tiny foramina are sometimes associated with these (Hutchison, 1991). Because the presence of musk glands does not necessarily result in the development of musk duct foramina, we decided to score these two characters separately. We relied on an unpublished thesis on the musk glands of turtles (Waagen, 1972) to determine the presence of musk glands for most taxa. In scoring taxa not investigated by Waagen (1972) we only recorded presence of musk glands if musk duct foramina provided positive evidence for their presence (e.g., baska, bealei, borneensis, galbinifrons, kachuga, mouhotii, petersi, pulcherrima, spengleri, and tentoria). Many tortoises, consequently, had to be scored as unknown, because they were not analyzed by Waagen (1972) and do not exhibit musk duct foramina (e.g., elongata, homeana, horsfieldi, and pardalis). Our scoring of the musk duct foramina is derived from a combination of osteological observation and data provided by Waagen (1972). Taxa not reported to possess musk glands (Waagen, 1972) were checked for musk duct foramina, but none were found. For those species that Waagen reported as having musk glands, we sought musk duct foramina on osteological specimens. Several taxa with musk glands, but only lightly ossified bridges, do not exhibit musk duct foramina (e.g., blandingii, dentata, flavomarginata, orbicularis, and pulcherrima) or show an asymmetry with foramina only visible anteriorly (e.g., N. platynota). In taxa that possess them, the musk duct foramina are sometimes contained entirely within the peripherals (e.g., N. platynota), and sometimes between the peripheral and the plastral buttress (e.g., reevesii). Polarity: Given the presence of musk glands in all extant outgroups (Waagen, 1972), their absence should be considered derived. Musk duct foramina are not described for lindholmemydids but J. H. Hutchison specifically searched for them in Mongolemys specimens housed at IVPP and found no trace of them. Because musk duct foramina are developed in the vast majority of the ingroup, we consider their absence to be derived for testudinoids. Plastron (59) Extensive overhanging lip of the epiplastra; 0 = absent, Fig. 120; 1 = present, Fig. 121 (Gaffney and Meylan 1988, H5.1, 9.2). In most testudinoid turtles, the epiplastra are rather flat with a slight increase in thickness along the anterior margin. In contrast, many tortoises have strongly thickened epiplastral margins that sometimes form an overhanging lip along the interior rim of the plastron. According to Gaffney and Meylan (1988) the presence of such an overhang of the epiplastra unites all tortoises except those classified in the genera Manouria (Fig. 120) and Gopherus, with a reversal occurring in giant insular forms. We confirm this general pattern, but we note the absence of an overhang in tornieri. A interior overhang is absent also in all extant emydids and batagurids, but a small overhang is present in extinct Ptychogaster and Echmatemys (Mlynarski, 1976, figs. 78, 81), taxa generally considered to belong to either the Emydidae or Batagurinae. Polarity: An overhanging lip on the epiplastra is absent in all outgroups and the majority of the ingroup. We consider its presence to be derived. (60) Intersection of the entoplastron by the humeropectoral sulcus; 0 = absent, Fig. 122; 1 = present, Fig. 123 (Hirayama, 1985, X; Crumly, 1985; Gaffney and Meylan, 1988, F5.1; McCord et al., 1995, 15). This character was used to help resolve relationships within batagurids by Hirayama (1985) but we were unable to replicate his results in our analysis. We agree with Crumly (1985) that the sulcus crosses the entoplastron in at least one species classified in the genus Indotestudo (i.e., elongata), but the sulcus is at the entoplastron/hyoplastron suture in our specimen of forsteni. The condition in species now commonly classified in Testudo varies widely (e.g., the suture crosses the entoplastron in graeca and horsfieldi, but does not in hermanni or kleinmanni). We also agree with Gaffney and Meylan (1988) on their distribution of this character among the Emydidae, however, our scoring for picta, orbicularis, and blandingii is polymorphic, because the sulcus generally runs along the suture between the entoplastron and the hypoplastra, but may barely fall on either side. Polarity: The polarity is ambiguous if only extant taxa are considered. The plastron of most outgroups is too different from that of testudinoids to be of any use for polarizing this character. For instance, the plastron of spinifera lacks scutes, and that of serpentina, odoratus, and caretta is too heavily modified to enable a meaningful comparison. Both character states are commonly found in batagurids, emydids, and testudinids, making an ingroup analysis futile. The humeropectoral sulcus is distinctly posterior to the epiplastron in Gravemys and Mongolemys, so we consider an intersection of this suture with the entoplastron to be derived.

19 2004 Asiatic Herpetological Research Vol. 10, p. 71 (61) Anal notch of the plastron; 0 = present, Fig. 124; 1 = greatly reduced, Fig. 125; 2 = absent, Fig. 126 (modified from Hirayama, 1985, W; Yasukawa et al., 2001, 22). The plastron of most testudinoid turtles has a significant anal notch. The absence of such an anal notch for amboinensis, galbinifrons, and flavomarginata was reported by Hirayama (1985) and we confirm those observations. An anal notch also is absent in belliana, carolina, coahuila, erosa, homeana, T. nelsoni, and T. ornata. To accommodate the presence of a reduced anal notch we modify Hirayama s (1985) character by creating a third character state. A reduced anal notch is found in blandingii, orbicularis, N. platynota, and reticularia. In at least one species (mouhotii), a distinct anal notch is present in larger individuals, but small specimens have a reduced notch (scored as polymorphic in our matrix); this suggests that development of an anal notch may be subject to ontogenetic variation in at least some testudinoids. Polarity: In caretta and serpentina an anal notch is not present, however, their plastra are narrow and tapered posteriorly. The fleshy plastron of spinifera is smooth along its posterior margin, but this cannot be observed in osteological preparations. A notch is weakly developed in at least some Lindholmemys, but is absent in Mongolemys. A notch is present in kinosternids, pleurodires, most of the ingroup, and in Gravemys. We conclude that its absence should be considered derived for our ingroup. (62) Anal scutes fused; 0 = absent, Fig. 127; 1 = present, Fig. 127 (Hirayama, 1985, Z). The anal scutes of adult galbinifrons, and flavomarginata are at least slightly fused, especially along their posterior medial border. We fully agree with Hirayama s (1985) treatment for this character. Anal scute fusion can be identified easily in macerated specimens (Fig. 127), because the anal scutes will not separate from one another, as will all other scutes. Polarity: Anal scute fusion is absent in the vast majority of turtles, and is considered to be the primitive condition. (63) Plastral scutes with vibrant, radiating color pattern; 0 = absent, Fig. 128; 1 = present, Fig. 129 (Hirayama, 1985, S; McCord et al., 1995, 16 Yasukawa et al., 2001, 32). Vibrant, radiating color patterns of the plastral scutes of dentata, grandis, and spinosa were noted by Hirayama (1985) and McCord et al. (1995). We add tcheponensis to this list, as well as the testudinids geometricus and P. oculifera. In our specimens, the pattern of dentata and tcheponensis is not as vibrant as in grandis and spinosa. Polarity: Vibrant, radiated color patterns are missing in all outgroups and the majority of the ingroup. Their presence is derived. Postcranium and Soft Tissue (64) Development of a suprascapula; 0 = absent; 1 = present, Fig. 130 (Gaffney and Meylan, 1988, F3.1; Burke et al., 1996, 11). (65) Development of an episcapula; 0 = absent; 1 = present, Fig. 130 (Gaffney and Meylan, 1988, F4.1; Burke et al., 1996, 11). The presence of both a suprascapula and an episcapula apparently is limited to blandingii and the species currently classified in Terrapene. A suprascapula is also present in orbicularis. Both structures are involved in the locking mechanism of the anterior plastral lobe during shell closure (Bramble, 1974). These structures are difficult to verify in most osteological preparations, because they may dissociate from the scapula and be difficult to recognize, and because they may ossify only in older individuals. The specimen we dissected to illustrate these features (TNHC 62532, a T. ornata with carapace length of 103 mm) has a completely cartilaginous episcapula, and a predominantly cartilaginous suprascapula (Fig. 130). It is therefore much easier to confirm their presence than verify their absence. We followed Bramble s (1974) account of these structures and scored our matrix accordingly, as probably did Gaffney and Meylan (1988) and Burke et al. (1996). Polarity: Suprascapulae and episcapulae are absent in all outgroups and the majority of the ingroup. Their presence is considered to be derived. (66) Shape of coracoid blade; 0 = long and narrow, Fig. 131; 1 = short and very wide, Fig. 131 (Crumly, 1985, 1994; Gaffney and Meylan, 1988, H1.7). The coracoid blade of all emydids and batagurids is an elongate bone with a narrow, short shaft and a long, wedge-shaped coracoid blade that is about two times wider than the base. In tortoises, this bone is still wedge-shaped, but relatively much shorter and with a blade that is considerably wider, typically four times the width of the base (Crumly, 1985, 1994; Gaffney and Meylan, 1988). We agree with previously published observations. Polarity: The coracoid blade of caretta, serpentina, and odoratus is long and narrow and that of spinifera is long, but not wedge-shaped. We consider a long and narrow coracoid blade to be primitive for Testudinoidea. (67) Number of manual claws; 0 = five, Fig. 132; 1 = four, Fig. 133 (modified from Hirayama, 1985, J).

20 Vol. 10, p. 72 Asiatic Herpetological Research 2004 Most testudinoid turtles have five manual claws with the exception of baska and horsfieldi, both of which have only four. We did not verify this character independently for all species, due to an overall lack of articulated skeletons and our limited access to pickled specimens. However, because the number of claws of the forelimbs is an easily determinable, discrete number that is regularly noted and described in the literature, we scored all remaining taxa from the comprehensive information provided by Ernst and Barbour (1989). Polarity: Five manual claws are present in serpentina, odoratus, and almost all pleurodires (excluding species currently classified in Chelodina and Hydromedusa); three are present in spinifera, and two in caretta. The condition in lindholmemydids is unknown. Based on ingroup commonality, we consider five claws to be primitive for the ingroup. (68) Number of phalanges of manus and pes; 0 = digital formula of or , Fig. 132; 1 = digital formula with less than , Fig. 133 (Crumly, 1985; Gaffney and Meylan, 1988, H1.1). The digital formula of most testudinoid turtles is or Among tortoises, the manus and pes are greatly shortened and the digital formula is typically reduced to or less (Auffenberg, 1974: ; Crumly, 1985). Due to the dissociated nature of most of the material we viewed, we were not able to verify the digital formulae of most of the turtle taxa we included. However, when articulated hands and feet were present, we never found anything to contradict the statements made above. We scored all tortoises based on information provided by Auffenberg (1974) and Crumly (1985). Polarity: All outgroups and the majority of the ingroup do not have a reduced digital formula. We consequently consider the reduced formula to be derived. (69) Webbing between digits; 0 = present, well developed, Fig. 134; 1 = absent, or at least strongly reduced, Fig. 135 (Hirayama, 1985, b). Due to their semi-aquatic nature, most testudinoids have well-developed webbing between the digits of their hands and feet. In more terrestrial species, however, the webbing often is reduced. Unfortunately, there seems to be a gradient in the development of webbing, from extremely well developed (e.g., baska, reticularia) to moderately developed (e.g., dentata, guttata) to virtually non-existent (e.g., spengleri). We nevertheless were able to reproduce Hirayama s (1985) distribution for the batagurids with the exception of grandis and spinosa, which have reduced webbing (grandis is the only borderline taxon we found, but its webbing is reduced relative to those taxa we scored as having well-developed webbing). Among emydids, we note that the webbing is heavily reduced in carolina, T. nelsoni, and T. ornata. All tortoises lack webbing. Polarity: All outgroups and the majority of ingroup taxa have webbed hands and feet. We consider the absence of webbing to be derived. (70) Sexual size dimorphism; 0 = absent; 1 = present, female much larger than male (Gaffney and Meylan, 1988, F5.2; Burke et al., 1996, 37). In almost every species of turtle, there is some expression of sexual size dimorphism (Berry and Shine, 1980; Gibbons and Lovich, 1990). The difference in size between the sexes can be expressed as a ratio and typically shows considerable variation depending on the population (Gibbons and Lovich, 1990). We initially tried to score this character with three character states, as done by Burke et al. (1996), differentiating between species with larger males, larger females, and equally sized sexes, but we abandoned that, because exact data are missing for most batagurid taxa. We consequently only score taxa as being sexually dimorphic if females are at least 1.4 times larger than the males. Our scores are derived from Gibbons and Lovich (1990) and Ernst and Barbour (1989). Polarity: Sexual size dimorphism is prevalent in most outgroups. In spinifera the female is much larger, in odoratus and serpentina the male tends to be slightly smaller, in caretta the sexes are of similar size. The outgroup polarity is thus ambiguous, but in the majority of the ingroup pronounced sexual dimorphism is absent. Problematic Characters We encountered difficulties in evaluating a number of previously used characters, and we provide some summary statements for those in this section. Most of these characters were not pursued thoroughly in our study because we were not able to understand the original descriptions, were unable recover discrete character states, or because at an early point in our investigation of the character we detected significant variation in expression of character states within taxa. In the latter case greater sample sizes or new methodological techniques (e.g., Wiens, 1995; Smith and Gutberlet, 2001) will be required to tease out a phylogenetic signal. (A) Frontal contribution to the supratemporal rim (Hirayama, 1985, 4). The anterior extent of the upper temporal emargination is difficult to define in many taxa, and is impossible to determine in those with a fully emarginated temporal region (e.g., T. ornata). The result is a high degree of ambiguity and a general lack of discrete character states.

21 2004 Asiatic Herpetological Research Vol. 10, p. 73 (B) Contact between postorbital and quadrate (Hirayama, 1985, 10). In the vast majority of ingroup taxa, there is no contact between the postorbital and the quadrate. Such a contact was observed only in japonica and punctularia by Hirayama (1985). In specimens of japonica available to us, we were not able to confirm this contact. CAS is a skeleton from a diseased specimen of punctularia. On the right side of the skull there is a possible (but only slight) contact. It is possible that the contact is actually between the postorbital and the quadratojugal (Fig. 136). We also found a minimal contact in one specimen of annulata. Given these diverging observations and the minute contact that is present in our material, we regard (for now) any contact within the ingroup as an abnormality. (C) Absence of the posterior process of the postorbital (Hirayama, 1985, 8). We cannot determine unambiguously what Hirayama (1985) meant by this character. In our assessment of testudinoids, both a posterolateral and a posteromedial process of the postorbital can occur. In Hirayama s (1985) analysis, only grandis and spinosa lack a posterior process of the postorbital. These species are also the only batagurid taxa to fully lack a temporal arch. We suspect that this character may somehow be referring to a lack of a bony temporal arch. (D) Processus inferior parietalis medially approximating each other, cranial cavity anteroventrally narrowing (Hirayama, 1985, 5; McCord et al., 1995, 7; Yasukawa et al., 2001, 2). We acknowledge the validity of this character as was originally worded by McDowell (1964). However, we find it difficult to determine how strongly the constriction of the brain case must be before it can be considered present. We were unable to develop unambiguous discrete character states for this feature. (E) Subdivision of the foramen nervi trigemini (Crumly, 1982; Hirayama, 1985, 6). This character was used originally by Crumly (1982) to infer phylogenetic relationships within Testudinidae. For his ingroup, Crumly (1982) observed a great amount of polymorphism, with no single species either completely lacking or always exhibiting a subdivision of the foramen. He also noted asymmetry for this character between the left and right side of some individuals. We confirm the common presence of a subdivided trigeminal foramen in representatives of Testudinidae, and the occasional presence in individuals of Emydidae and Bataguridae (e.g., areolata, dentata, flavimaculata, N. platynota, and rubida). A significant amount of variation can be observed in two specimens of borneensis available to us, that exhibit left/right asymmetry and the full spectrum from a fully intact (Fig. 137), to partially subdivided (Fig. 138), to fully subdivided (Fig. 139) trigeminal foramen. Given that most taxa are represented by three or fewer skulls in our study, it is apparent that we are not able to fully document the amount of variation exhibited by testudinoids. (F) Contact between postorbital and squamosal (Hirayama, 1985, 9). Gaffney et al. (1991) noted that absence of this contact is associated with the upper temporal emargination and considered it informative at the level of their analysis. Within our ingroup, all turtles have substantial upper temporal emarginations, resulting in the contact being just barely present, or just barely absent, or polymorphic (e.g., picta, petersi, texana, crassicollis). See comments above under character 9. (G) Median premaxillary notch (Hirayama, 1985, 19; Yasukawa et al., 2001, 9) (H) Large cusps near the suture of the premaxillae and maxillae (Hirayama, 1985, 28). Initially, we were faced with the problem of defining these two characters independently from one another, because the presence of two tightly spaced, opposing cusp-like structures along any margin will automatically result in the development of a median notch. An additional problem relating to these characters is the question of whether these features should be observed on the ramphotheca or the maxilla. Large, tooth-like cusps are clearly present in a number of taxa (e.g., thurjii) but so is the full spectrum of smaller cusps, making it impossible to clearly define discrete character states. Furthermore, if all species were evaluated for medial notches that existed even if the cusps were removed, all taxa in our sample would show a medial notch. We were unable to develop a consistent method for scoring this character for all testudinoid species. (I) Antero-medial portion of the upper triturating surface formed by premaxillae and maxillae (Hirayama, 1985, 23; Yasukawa et al., 2001, 13). We are neither able to replicate the full meaning of this character nor formulate truly discrete character states. A connection with the development of the secondary palate is evident, but the morphology of this region seems to be sufficiently covered by a number of other characters.

22 Vol. 10, p. 74 Asiatic Herpetological Research 2004 (J) Participation of the vomer in the foramen praepalatinum (Crumly, 1982, 10; Hirayama, 1985, 32; Yasukawa et al., 2001, 15). Within testudinoid turtles, the foramen praepalatinum perforates the nasal cavity at the border between the premaxilla and the vomer. When the foramen is positioned slightly more anteriorly, it is fully surrounded by the premaxilla, when it is minutely farther posterior it is surrounded by the vomer. Considering the impact of such minute changes, it is not surprising that our scoring for this character generally seems to be in conflict with that of Hirayama (1985) and Crumly (1982). This character appears to be subject to great intraspecific variation. (K) Foramen palatinum posterius enclosed within the brain cavity (Hirayama, 1985, 34). According to Hirayama (1985), in reevesii (only) the foramen palatinum posterius is enclosed within the region of the brain cavity due to a flared descending process of the parietal. We cannot confirm this observation for any testudinoid turtles (including three specimens of reevesii). (L) Participating bones in the processus trochlearis oticum (Hirayama 1985, 37; Gaffney and Meylan, 1988, Gaffney et al. 1991, 6; McCord et al., 1995, 8; Shaffer et al., 1997, 74, 258 Yasukawa et al., 2001, 18). The relative participation of the prootic, parietal, and quadrate to the processus trochlearis oticum was used previously by a number of authors to infer phylogenetic relationships within turtles. Our observations confirm the great variety of morphologies that can be observed in this region. However, the amount of intraspecific variation is considerable and the full spectrum of possible morphologies seems to be filled, making it difficult to discern discrete character states. Future research in the area may result in more clearly defined discrete character states. (M) Length of the crista supraoccipitalis (Hirayama, 1985, 40). A long crista supraoccipitalis was observed by Hirayama (1985) for borneensis. The character states he used are defined by relative length of the crista supraoccipitalis to the condylo-basal length. Unfortunately, we could not replicate this because it is not clear exactly how the length of the crista was measured. Furthermore, a true morphological gap seems to be missing between the admittedly very long crista of borneensis and other batagurids with an elongated crista. This character is problematic, because it is poorly defined and lacks discrete character states. (N) Bony sutures and sulci lost in old adults (Hirayama, 1985, I). According to Hirayama (1985), loss of sutures and sulci occurs in baska, borneoensis, and borneensis only. We are able to confirm this, but we do not have individuals of all other species that are sufficiently old enough to positively confirm that they also do not exhibit this feature at old age. In subsequent treatments, loss of sutures and loss of sulci should be treated as separate characters. (O) Ossification of cornu branchiale II (Hirayama, 1985, 48; Yasukawa et al., 2001, 20). This character was used previously to unite tortoises with a number of batagurid taxa (Hirayama, 1985). The hyoid apparatus of turtles is often disarticulated in skeletal preparations, making is difficult to positively confirm if an ossified cornu branchiale is present or absent. However, for those taxa for which we were able to observe the hyoid apparatus, we were not able to confirm Hirayama s (1985) observation of a reduced cornu branchiale II in some batagurids (e.g., galbinifrons, spengleri). Instead, these taxa exhibit a cornu branchiale II that is not significantly different from most other batagurids. (P) Double articulation between the fifth and sixth cervical vertebrae (Hirayama, 1985; Gaffney and Meylan, 1988, F1.5). Most articular surfaces of the cervical column are rather homogenous within all testudinoid turtles (Williams, 1950). A double articulation between the fifth and sixth cervical previously was considered to be a unique character that unites the Emydinae (McDowell, 1964). This character also was used by Hirayama (1985) and with reservations by Gaffney and Meylan (1988). Our observations generally confirm the presence of a more or less clear double articulation in most emydids, however, this features is also present in a number of batagurids confirming that this character is highly variable within the ingroup (Williams, 1950; Gaffney and Meylan, 1988). Unfortunately, discrete character states are lacking; we were able to observe the full morphological spectrum from a clear singular articulation to a clear double articulation. (Q) Scapular prong with lateral concavity (Hirayama, 1985, E). Hirayama (1985) reported this character as an autapomorphy for subtrijuga only. However, we cannot identify this morphology in any of our specimens of subtrijuga.

23 2004 Asiatic Herpetological Research Vol. 10, p. 75 (R) Large facet of the ilium for the testoscapularis and testoiliacus (Hirayama, 1985, T; Yasukawa et al., 2001, 34) (S) Extensive development of both testoscapularis and testoiliacus (Hirayama, 1985, U). An extensive development of the testoscapularis and testoiliacus muscles together with an associated large scar on the ilium was reported by Bramble (1974) for Asian and North American box turtles. Whereas we have no reason to doubt his assessment of the development of these muscles for box turtles, we were not able to score this character for most of the remaining taxa. The shape of the ilium was explored and illustrated for some emydids and batagurids by Yasukawa et al. (2001: ). (T) Ossification of the epipubis (Gaffney and Meylan, 1988, F5.5). The identification (and confirmation of presence or absence) of an ossified epipubis (Fig. 141) is somewhat difficult for most species, because it seems to ossify rather late in ontogeny, and can fall off during preparation. Our tentative observations confirm the presence of an ossified epipubis in numerous adult emydids and batagurids, typically terrestrial forms (e.g., T. ornata, G. insculpta, N. platynota, yuwonoi). An improved sample of adult specimens of all taxa, however, is necessary to reveal the true distribution of this character. (U) Diploid Number of Chromosomes (Hirayama, 1985, Ø; Shaffer et al., 1997, 43). The diploid number of chromosomes was used by Carr and Bickham (1986) to hypothesize a sister group relationship between the Emydinae and subtrijuga, followed by borneensis and crassicollis and finally the rest of the Batagurinae. Whereas most batagurids allegedly have 52 chromosomes, subtrijuga, borneensis, crassicollis, and emydids are supposed to have 50 chromosomes. We view these results with caution, because a brief review of the relevant literature reveals great differences in chromosomal counts for a variety of taxa. For instance, according to the work of Killebrew (1977) and Bickham (1981), amboinensis has 52 chromosomes, however, Gorman (1973) reported only 50. Similar conflicts can also be found for dentata (Bickham, 1981; DeSmet, 1978; Gorman, 1973; Stock, 1972), subtrijuga (Bickham, 1981; Killebrew, 1977), trijuga (DeSmet, 1978; Carr and Bickham, 1986), sinensis (Bickham, 1981; Killebrew, 1977; Stock, 1972), crassicollis (Killebrew, 1977; Stock, 1972; Bickham and Baker, 1976), and some of the species currently placed in Graptemys (Killebrew, 1977; McKown 1972), Trachemys (DeSmet, 1978; Killebrew, 1977; Stock, 1972) and Mauremys (Killebrew, 1977; Stock, 1972). This conflict in primary data is probably best understood when considering the nature of testudinoid chromosomes: whereas 14 pairs of chromosomes have a considerable size, all of the remaining ones are extremely small. Given these circumstances, it seems reasonable to hypothesize that one pair of chromosomes may be unrecognized during analysis. (V) Plica media spade-shaped (Gaffney and Meylan, 1988, F7.1). The penile soft anatomy of turtles was comprehensively reviewed by Zug (1966) and one of his characters, the shape of the plica media, was used by Gaffney and Meylan (1988) to unify species placed in Chrysemys, Deirochelys, Trachemys, and Pseudemys as a monophyletic group. In his detailed description of the plica media, Zug (1966) referred to the shape of this structure as being spade-shaped in those taxa, but made similar claims for other taxa too. Furthermore, based on the illustrations that were provided by Zug (1966) for other taxa, the plica media of species placed in Graptemys, Malaclemys, Rhinoclemmys, and Platysternon appear spade-shaped also, even though Zug (1966) did not explicitly use those descriptive words. This anatomical system should be carefully reevaluated for all testudinoids, with special attention given to definition of discrete characters. (W) Ossifications within the fenestra postotica. In some taxa, portions of the fenestra postotica are closed or obscured by ossifications (noted, but without exemplars, by Gaffney, 1972). In our largest specimen of grandis (CAS ) a short, spike-like ossified process extends posterodorsally from the dorsal edge of the quadrate process of the pterygoid, and crosses the fenestra postotica. It is situated ventral to the stapes (columella auris), medial to the incisura columella auris, and lateral to the fenestra ovalis (Fig. 141). In some specimens, the dorsal tip of a similar structure approaches or meets a posterodorsally-inclined process that extends from the dorsal surface of the pterygoid, near the suture with the prootic. In our largest specimen of N. platynota (CAS ) the two processes meet to enclose the stapedial shaft in a ring of bone situated at approximately the midpoint between the fenestra ovalis and the medial opening of the incisura columella auris. Our other, smaller, specimen of N. platynota shows no development of these processes. It seems likely that there is an ontogenetic component in the expression of this feature. It does not appear to have any systematic significance, but in any case it is not widespread within Testudinoidea.

24 Vol. 10, p. 76 Asiatic Herpetological Research 2004 Conclusions Our observations can be used to draw some tentative conclusions regarding the current level of understanding about morphological variation within testudinoid turtles. In addition, we provide some cautionary statements about the quality of morphological data now in use for assessing systematic relationships among these turtles. It is clear from a perusal of the relevant literature, and from our data, that there is reasonably strong morphological support for a monophyletic Testudinidae. Support for monophyly of emydines and batagurines is not as impressive. The paraphyly of Batagurinae (with respect to Testudinidae ) was explicitly proposed by Hirayama (1985) and has been generally accepted since that time, although some strides have been made towards resolving relationships among some batagurine taxa. The monophyly of Emydinae seems to have been implicitly assumed by many workers, but remains to be established in the context of a rigorous phylogenetic analysis of all relevant taxa. The interpretation of several morphological features shared between some emydines and some batagurines as either convergence or synapomorphy remains an important and interesting challenge. For example, there are intriguing morphological similarities between subtrijuga and some species classified in the genus Graptemys (e.g., contact of the jugal and descending process of the parietal; contact of the quadratojugal with the articular facet of the quadrate; contact between the quadratojugal and the maxilla; ventral process of the pterygoid approaching the articular surface of the quadrate). These similarities may be due to functional convergence as a result of a molluscivorous diet, but they raise questions about the propriety of utilizing subtrijuga as an outgroup for systematic studies of emydines (e.g., Burke et al., 1996). Additional similarities are reported for chromosome numbers in subtrijuga and emydines (see Problematic Character U, above). A seriously deficient understanding of morphological variation is one of the greatest inadequacies of current perspectives on morphological data in turtles generally, and testudinoids specifically. Few published studies have been conducted to evaluate the range and causes of sexual, ontogenetic, intra- and inter-population variation in morphological characters within testudinoids. Our preliminary considerations of ontogenetic variation, combined with reports of sexual variation (e.g., Berry and Shine 1980; Gibbons and Lovich, 1990; Stephens, 1998) and new studies exploring the complex interactions of morphological evolution with behavioral characteristics and environmental conditions in turtles (e.g., Lindeman, 2000; Herrel et al., 2002; Joyce and Gauthier, 2003; Claude et al., 2003) emphasize the importance of pursuing these questions further. Our decision not to produce a phylogenetic hypothesis in this paper was based primarily on two considerations. The first is the relatively small sample size we used for many taxa in this study (although it is comparable to sample sizes from other, previously published, studies), and the fact that several taxa are not represented in our work. The second consideration is our sense that the current understanding of morphological variation in testudinoid turtles is insufficiently mature to permit reliable phylogenetic hypotheses based solely on morphological data. The most expedient way to address the need for greater documentation of variation within testudinoid species is to utilize existing museum collections to the greatest extent possible, and secondarily to develop responsible collecting programs that are designed with this need in mind. Our results also indicate that morphological data matrices currently in the literature should not be taken at face value. We had particular difficulties replicating some of the scoring in the Hirayama (1985) matrix. That seminal analysis (completed prior to the widespread use of computer-assisted analytical methods in systematics) laid the foundation for nearly all subsequent work on batagurine morphology (including our own), and its importance in shaping our current conceptualization of batagurine phylogeny cannot be overstated. The work of pathfinders in all fields of inquiry is often subjected to the greatest scrutiny by the next generation of researchers. Our statements and contradictory observations in this paper in no way denigrate Hirayama s work; instead, we view our efforts as minor attempts to correct the few inconsistencies in his analysis, and to contribute our observations to the body of knowledge that he began to synthesize 20 years ago. The accurate interpretation and understanding of the morphological descriptions of previous authors were among the great challenges we faced when we began our studies of testudinoid skeletal morphology. Much of our confusion could have been averted if adequate illustrations accompanied all published character descriptions, but such documentation often is an expensive undertaking. Our photographs of character states discussed in this paper are intended to facilitate communication among turtle enthusiasts, and to provide a baseline for future comparisons and discussions about testudinoid morphology. We hope that adequate illustration of all newly proposed characters will become standard practice among turtle systematists. It seems likely that our interpretations of characters will differ in some respects from those of our colleagues, and we anticipate that our decisions regarding problematic characters (discussed above), and our choices with respect to lumping or splitting previously published character states, will generate much spirited discussion in the years ahead.

25 2004 Asiatic Herpetological Research Vol. 10, p. 77

26 Vol. 10, p. 78 Asiatic Herpetological Research 2004

27 2004 Asiatic Herpetological Research Vol. 10, p. 79

28 Vol. 10, p. 80 Asiatic Herpetological Research 2004

29 2004 Asiatic Herpetological Research Vol. 10, p. 81

30 Vol. 10, p. 82 Asiatic Herpetological Research 2004

31 2004 Asiatic Herpetological Research Vol. 10, p. 83

32 Vol. 10, p. 84 Asiatic Herpetological Research 2004

33 2004 Asiatic Herpetological Research Vol. 10, p. 85

34 Vol. 10, p. 86 Asiatic Herpetological Research 2004

35 2004 Asiatic Herpetological Research Vol. 10, p. 87

36 Vol. 10, p. 88 Asiatic Herpetological Research 2004

37 2004 Asiatic Herpetological Research Vol. 10, p. 89

38 Vol. 10, p. 90 Asiatic Herpetological Research 2004

39 2004 Asiatic Herpetological Research Vol. 10, p. 91

40 Vol. 10, p. 92 Asiatic Herpetological Research 2004

41 2004 Asiatic Herpetological Research Vol. 10, p. 93

42 Vol. 10, p. 94 Asiatic Herpetological Research 2004

43 2004 Asiatic Herpetological Research Vol. 10, p. 95

44 Vol. 10, p. 96 Asiatic Herpetological Research 2004

45 2004 Asiatic Herpetological Research Vol. 10, p. 97

46 Vol. 10, p. 98 Asiatic Herpetological Research 2004

47 2004 Asiatic Herpetological Research Vol. 10, p. 99

48 Vol. 10, p. 100 Asiatic Herpetological Research 2004